OpenCloudOS-Kernel/drivers/md/dm-cache-policy-smq.c

1943 lines
44 KiB
C
Raw Normal View History

dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
/*
* Copyright (C) 2015 Red Hat. All rights reserved.
*
* This file is released under the GPL.
*/
#include "dm-cache-background-tracker.h"
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
#include "dm-cache-policy-internal.h"
#include "dm-cache-policy.h"
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
#include "dm.h"
#include <linux/hash.h>
#include <linux/jiffies.h>
#include <linux/module.h>
#include <linux/mutex.h>
#include <linux/vmalloc.h>
#include <linux/math64.h>
#define DM_MSG_PREFIX "cache-policy-smq"
/*----------------------------------------------------------------*/
/*
* Safe division functions that return zero on divide by zero.
*/
static unsigned safe_div(unsigned n, unsigned d)
{
return d ? n / d : 0u;
}
static unsigned safe_mod(unsigned n, unsigned d)
{
return d ? n % d : 0u;
}
/*----------------------------------------------------------------*/
struct entry {
unsigned hash_next:28;
unsigned prev:28;
unsigned next:28;
unsigned level:6;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
bool dirty:1;
bool allocated:1;
bool sentinel:1;
bool pending_work:1;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
dm_oblock_t oblock;
};
/*----------------------------------------------------------------*/
#define INDEXER_NULL ((1u << 28u) - 1u)
/*
* An entry_space manages a set of entries that we use for the queues.
* The clean and dirty queues share entries, so this object is separate
* from the queue itself.
*/
struct entry_space {
struct entry *begin;
struct entry *end;
};
static int space_init(struct entry_space *es, unsigned nr_entries)
{
if (!nr_entries) {
es->begin = es->end = NULL;
return 0;
}
treewide: Use array_size() in vzalloc() The vzalloc() function has no 2-factor argument form, so multiplication factors need to be wrapped in array_size(). This patch replaces cases of: vzalloc(a * b) with: vzalloc(array_size(a, b)) as well as handling cases of: vzalloc(a * b * c) with: vzalloc(array3_size(a, b, c)) This does, however, attempt to ignore constant size factors like: vzalloc(4 * 1024) though any constants defined via macros get caught up in the conversion. Any factors with a sizeof() of "unsigned char", "char", and "u8" were dropped, since they're redundant. The Coccinelle script used for this was: // Fix redundant parens around sizeof(). @@ type TYPE; expression THING, E; @@ ( vzalloc( - (sizeof(TYPE)) * E + sizeof(TYPE) * E , ...) | vzalloc( - (sizeof(THING)) * E + sizeof(THING) * E , ...) ) // Drop single-byte sizes and redundant parens. @@ expression COUNT; typedef u8; typedef __u8; @@ ( vzalloc( - sizeof(u8) * (COUNT) + COUNT , ...) | vzalloc( - sizeof(__u8) * (COUNT) + COUNT , ...) | vzalloc( - sizeof(char) * (COUNT) + COUNT , ...) | vzalloc( - sizeof(unsigned char) * (COUNT) + COUNT , ...) | vzalloc( - sizeof(u8) * COUNT + COUNT , ...) | vzalloc( - sizeof(__u8) * COUNT + COUNT , ...) | vzalloc( - sizeof(char) * COUNT + COUNT , ...) | vzalloc( - sizeof(unsigned char) * COUNT + COUNT , ...) ) // 2-factor product with sizeof(type/expression) and identifier or constant. @@ type TYPE; expression THING; identifier COUNT_ID; constant COUNT_CONST; @@ ( vzalloc( - sizeof(TYPE) * (COUNT_ID) + array_size(COUNT_ID, sizeof(TYPE)) , ...) | vzalloc( - sizeof(TYPE) * COUNT_ID + array_size(COUNT_ID, sizeof(TYPE)) , ...) | vzalloc( - sizeof(TYPE) * (COUNT_CONST) + array_size(COUNT_CONST, sizeof(TYPE)) , ...) | vzalloc( - sizeof(TYPE) * COUNT_CONST + array_size(COUNT_CONST, sizeof(TYPE)) , ...) | vzalloc( - sizeof(THING) * (COUNT_ID) + array_size(COUNT_ID, sizeof(THING)) , ...) | vzalloc( - sizeof(THING) * COUNT_ID + array_size(COUNT_ID, sizeof(THING)) , ...) | vzalloc( - sizeof(THING) * (COUNT_CONST) + array_size(COUNT_CONST, sizeof(THING)) , ...) | vzalloc( - sizeof(THING) * COUNT_CONST + array_size(COUNT_CONST, sizeof(THING)) , ...) ) // 2-factor product, only identifiers. @@ identifier SIZE, COUNT; @@ vzalloc( - SIZE * COUNT + array_size(COUNT, SIZE) , ...) // 3-factor product with 1 sizeof(type) or sizeof(expression), with // redundant parens removed. @@ expression THING; identifier STRIDE, COUNT; type TYPE; @@ ( vzalloc( - sizeof(TYPE) * (COUNT) * (STRIDE) + array3_size(COUNT, STRIDE, sizeof(TYPE)) , ...) | vzalloc( - sizeof(TYPE) * (COUNT) * STRIDE + array3_size(COUNT, STRIDE, sizeof(TYPE)) , ...) | vzalloc( - sizeof(TYPE) * COUNT * (STRIDE) + array3_size(COUNT, STRIDE, sizeof(TYPE)) , ...) | vzalloc( - sizeof(TYPE) * COUNT * STRIDE + array3_size(COUNT, STRIDE, sizeof(TYPE)) , ...) | vzalloc( - sizeof(THING) * (COUNT) * (STRIDE) + array3_size(COUNT, STRIDE, sizeof(THING)) , ...) | vzalloc( - sizeof(THING) * (COUNT) * STRIDE + array3_size(COUNT, STRIDE, sizeof(THING)) , ...) | vzalloc( - sizeof(THING) * COUNT * (STRIDE) + array3_size(COUNT, STRIDE, sizeof(THING)) , ...) | vzalloc( - sizeof(THING) * COUNT * STRIDE + array3_size(COUNT, STRIDE, sizeof(THING)) , ...) ) // 3-factor product with 2 sizeof(variable), with redundant parens removed. @@ expression THING1, THING2; identifier COUNT; type TYPE1, TYPE2; @@ ( vzalloc( - sizeof(TYPE1) * sizeof(TYPE2) * COUNT + array3_size(COUNT, sizeof(TYPE1), sizeof(TYPE2)) , ...) | vzalloc( - sizeof(TYPE1) * sizeof(THING2) * (COUNT) + array3_size(COUNT, sizeof(TYPE1), sizeof(TYPE2)) , ...) | vzalloc( - sizeof(THING1) * sizeof(THING2) * COUNT + array3_size(COUNT, sizeof(THING1), sizeof(THING2)) , ...) | vzalloc( - sizeof(THING1) * sizeof(THING2) * (COUNT) + array3_size(COUNT, sizeof(THING1), sizeof(THING2)) , ...) | vzalloc( - sizeof(TYPE1) * sizeof(THING2) * COUNT + array3_size(COUNT, sizeof(TYPE1), sizeof(THING2)) , ...) | vzalloc( - sizeof(TYPE1) * sizeof(THING2) * (COUNT) + array3_size(COUNT, sizeof(TYPE1), sizeof(THING2)) , ...) ) // 3-factor product, only identifiers, with redundant parens removed. @@ identifier STRIDE, SIZE, COUNT; @@ ( vzalloc( - (COUNT) * STRIDE * SIZE + array3_size(COUNT, STRIDE, SIZE) , ...) | vzalloc( - COUNT * (STRIDE) * SIZE + array3_size(COUNT, STRIDE, SIZE) , ...) | vzalloc( - COUNT * STRIDE * (SIZE) + array3_size(COUNT, STRIDE, SIZE) , ...) | vzalloc( - (COUNT) * (STRIDE) * SIZE + array3_size(COUNT, STRIDE, SIZE) , ...) | vzalloc( - COUNT * (STRIDE) * (SIZE) + array3_size(COUNT, STRIDE, SIZE) , ...) | vzalloc( - (COUNT) * STRIDE * (SIZE) + array3_size(COUNT, STRIDE, SIZE) , ...) | vzalloc( - (COUNT) * (STRIDE) * (SIZE) + array3_size(COUNT, STRIDE, SIZE) , ...) | vzalloc( - COUNT * STRIDE * SIZE + array3_size(COUNT, STRIDE, SIZE) , ...) ) // Any remaining multi-factor products, first at least 3-factor products // when they're not all constants... @@ expression E1, E2, E3; constant C1, C2, C3; @@ ( vzalloc(C1 * C2 * C3, ...) | vzalloc( - E1 * E2 * E3 + array3_size(E1, E2, E3) , ...) ) // And then all remaining 2 factors products when they're not all constants. @@ expression E1, E2; constant C1, C2; @@ ( vzalloc(C1 * C2, ...) | vzalloc( - E1 * E2 + array_size(E1, E2) , ...) ) Signed-off-by: Kees Cook <keescook@chromium.org>
2018-06-13 05:27:37 +08:00
es->begin = vzalloc(array_size(nr_entries, sizeof(struct entry)));
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
if (!es->begin)
return -ENOMEM;
es->end = es->begin + nr_entries;
return 0;
}
static void space_exit(struct entry_space *es)
{
vfree(es->begin);
}
static struct entry *__get_entry(struct entry_space *es, unsigned block)
{
struct entry *e;
e = es->begin + block;
BUG_ON(e >= es->end);
return e;
}
static unsigned to_index(struct entry_space *es, struct entry *e)
{
BUG_ON(e < es->begin || e >= es->end);
return e - es->begin;
}
static struct entry *to_entry(struct entry_space *es, unsigned block)
{
if (block == INDEXER_NULL)
return NULL;
return __get_entry(es, block);
}
/*----------------------------------------------------------------*/
struct ilist {
unsigned nr_elts; /* excluding sentinel entries */
unsigned head, tail;
};
static void l_init(struct ilist *l)
{
l->nr_elts = 0;
l->head = l->tail = INDEXER_NULL;
}
static struct entry *l_head(struct entry_space *es, struct ilist *l)
{
return to_entry(es, l->head);
}
static struct entry *l_tail(struct entry_space *es, struct ilist *l)
{
return to_entry(es, l->tail);
}
static struct entry *l_next(struct entry_space *es, struct entry *e)
{
return to_entry(es, e->next);
}
static struct entry *l_prev(struct entry_space *es, struct entry *e)
{
return to_entry(es, e->prev);
}
static bool l_empty(struct ilist *l)
{
return l->head == INDEXER_NULL;
}
static void l_add_head(struct entry_space *es, struct ilist *l, struct entry *e)
{
struct entry *head = l_head(es, l);
e->next = l->head;
e->prev = INDEXER_NULL;
if (head)
head->prev = l->head = to_index(es, e);
else
l->head = l->tail = to_index(es, e);
if (!e->sentinel)
l->nr_elts++;
}
static void l_add_tail(struct entry_space *es, struct ilist *l, struct entry *e)
{
struct entry *tail = l_tail(es, l);
e->next = INDEXER_NULL;
e->prev = l->tail;
if (tail)
tail->next = l->tail = to_index(es, e);
else
l->head = l->tail = to_index(es, e);
if (!e->sentinel)
l->nr_elts++;
}
static void l_add_before(struct entry_space *es, struct ilist *l,
struct entry *old, struct entry *e)
{
struct entry *prev = l_prev(es, old);
if (!prev)
l_add_head(es, l, e);
else {
e->prev = old->prev;
e->next = to_index(es, old);
prev->next = old->prev = to_index(es, e);
if (!e->sentinel)
l->nr_elts++;
}
}
static void l_del(struct entry_space *es, struct ilist *l, struct entry *e)
{
struct entry *prev = l_prev(es, e);
struct entry *next = l_next(es, e);
if (prev)
prev->next = e->next;
else
l->head = e->next;
if (next)
next->prev = e->prev;
else
l->tail = e->prev;
if (!e->sentinel)
l->nr_elts--;
}
static struct entry *l_pop_head(struct entry_space *es, struct ilist *l)
{
struct entry *e;
for (e = l_head(es, l); e; e = l_next(es, e))
if (!e->sentinel) {
l_del(es, l, e);
return e;
}
return NULL;
}
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
static struct entry *l_pop_tail(struct entry_space *es, struct ilist *l)
{
struct entry *e;
for (e = l_tail(es, l); e; e = l_prev(es, e))
if (!e->sentinel) {
l_del(es, l, e);
return e;
}
return NULL;
}
/*----------------------------------------------------------------*/
/*
* The stochastic-multi-queue is a set of lru lists stacked into levels.
* Entries are moved up levels when they are used, which loosely orders the
* most accessed entries in the top levels and least in the bottom. This
* structure is *much* better than a single lru list.
*/
#define MAX_LEVELS 64u
struct queue {
struct entry_space *es;
unsigned nr_elts;
unsigned nr_levels;
struct ilist qs[MAX_LEVELS];
/*
* We maintain a count of the number of entries we would like in each
* level.
*/
unsigned last_target_nr_elts;
unsigned nr_top_levels;
unsigned nr_in_top_levels;
unsigned target_count[MAX_LEVELS];
};
static void q_init(struct queue *q, struct entry_space *es, unsigned nr_levels)
{
unsigned i;
q->es = es;
q->nr_elts = 0;
q->nr_levels = nr_levels;
for (i = 0; i < q->nr_levels; i++) {
l_init(q->qs + i);
q->target_count[i] = 0u;
}
q->last_target_nr_elts = 0u;
q->nr_top_levels = 0u;
q->nr_in_top_levels = 0u;
}
static unsigned q_size(struct queue *q)
{
return q->nr_elts;
}
/*
* Insert an entry to the back of the given level.
*/
static void q_push(struct queue *q, struct entry *e)
{
BUG_ON(e->pending_work);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
if (!e->sentinel)
q->nr_elts++;
l_add_tail(q->es, q->qs + e->level, e);
}
static void q_push_front(struct queue *q, struct entry *e)
{
BUG_ON(e->pending_work);
if (!e->sentinel)
q->nr_elts++;
l_add_head(q->es, q->qs + e->level, e);
}
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
static void q_push_before(struct queue *q, struct entry *old, struct entry *e)
{
BUG_ON(e->pending_work);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
if (!e->sentinel)
q->nr_elts++;
l_add_before(q->es, q->qs + e->level, old, e);
}
static void q_del(struct queue *q, struct entry *e)
{
l_del(q->es, q->qs + e->level, e);
if (!e->sentinel)
q->nr_elts--;
}
/*
* Return the oldest entry of the lowest populated level.
*/
static struct entry *q_peek(struct queue *q, unsigned max_level, bool can_cross_sentinel)
{
unsigned level;
struct entry *e;
max_level = min(max_level, q->nr_levels);
for (level = 0; level < max_level; level++)
for (e = l_head(q->es, q->qs + level); e; e = l_next(q->es, e)) {
if (e->sentinel) {
if (can_cross_sentinel)
continue;
else
break;
}
return e;
}
return NULL;
}
static struct entry *q_pop(struct queue *q)
{
struct entry *e = q_peek(q, q->nr_levels, true);
if (e)
q_del(q, e);
return e;
}
/*
* This function assumes there is a non-sentinel entry to pop. It's only
* used by redistribute, so we know this is true. It also doesn't adjust
* the q->nr_elts count.
*/
static struct entry *__redist_pop_from(struct queue *q, unsigned level)
{
struct entry *e;
for (; level < q->nr_levels; level++)
for (e = l_head(q->es, q->qs + level); e; e = l_next(q->es, e))
if (!e->sentinel) {
l_del(q->es, q->qs + e->level, e);
return e;
}
return NULL;
}
static void q_set_targets_subrange_(struct queue *q, unsigned nr_elts, unsigned lbegin, unsigned lend)
{
unsigned level, nr_levels, entries_per_level, remainder;
BUG_ON(lbegin > lend);
BUG_ON(lend > q->nr_levels);
nr_levels = lend - lbegin;
entries_per_level = safe_div(nr_elts, nr_levels);
remainder = safe_mod(nr_elts, nr_levels);
for (level = lbegin; level < lend; level++)
q->target_count[level] =
(level < (lbegin + remainder)) ? entries_per_level + 1u : entries_per_level;
}
/*
* Typically we have fewer elements in the top few levels which allows us
* to adjust the promote threshold nicely.
*/
static void q_set_targets(struct queue *q)
{
if (q->last_target_nr_elts == q->nr_elts)
return;
q->last_target_nr_elts = q->nr_elts;
if (q->nr_top_levels > q->nr_levels)
q_set_targets_subrange_(q, q->nr_elts, 0, q->nr_levels);
else {
q_set_targets_subrange_(q, q->nr_in_top_levels,
q->nr_levels - q->nr_top_levels, q->nr_levels);
if (q->nr_in_top_levels < q->nr_elts)
q_set_targets_subrange_(q, q->nr_elts - q->nr_in_top_levels,
0, q->nr_levels - q->nr_top_levels);
else
q_set_targets_subrange_(q, 0, 0, q->nr_levels - q->nr_top_levels);
}
}
static void q_redistribute(struct queue *q)
{
unsigned target, level;
struct ilist *l, *l_above;
struct entry *e;
q_set_targets(q);
for (level = 0u; level < q->nr_levels - 1u; level++) {
l = q->qs + level;
target = q->target_count[level];
/*
* Pull down some entries from the level above.
*/
while (l->nr_elts < target) {
e = __redist_pop_from(q, level + 1u);
if (!e) {
/* bug in nr_elts */
break;
}
e->level = level;
l_add_tail(q->es, l, e);
}
/*
* Push some entries up.
*/
l_above = q->qs + level + 1u;
while (l->nr_elts > target) {
e = l_pop_tail(q->es, l);
if (!e)
/* bug in nr_elts */
break;
e->level = level + 1u;
l_add_tail(q->es, l_above, e);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
}
}
}
static void q_requeue(struct queue *q, struct entry *e, unsigned extra_levels,
struct entry *s1, struct entry *s2)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
struct entry *de;
unsigned sentinels_passed = 0;
unsigned new_level = min(q->nr_levels - 1u, e->level + extra_levels);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
/* try and find an entry to swap with */
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
if (extra_levels && (e->level < q->nr_levels - 1u)) {
for (de = l_head(q->es, q->qs + new_level); de && de->sentinel; de = l_next(q->es, de))
sentinels_passed++;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
if (de) {
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
q_del(q, de);
de->level = e->level;
if (s1) {
switch (sentinels_passed) {
case 0:
q_push_before(q, s1, de);
break;
case 1:
q_push_before(q, s2, de);
break;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
default:
q_push(q, de);
}
} else
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
q_push(q, de);
}
}
q_del(q, e);
e->level = new_level;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
q_push(q, e);
}
/*----------------------------------------------------------------*/
#define FP_SHIFT 8
#define SIXTEENTH (1u << (FP_SHIFT - 4u))
#define EIGHTH (1u << (FP_SHIFT - 3u))
struct stats {
unsigned hit_threshold;
unsigned hits;
unsigned misses;
};
enum performance {
Q_POOR,
Q_FAIR,
Q_WELL
};
static void stats_init(struct stats *s, unsigned nr_levels)
{
s->hit_threshold = (nr_levels * 3u) / 4u;
s->hits = 0u;
s->misses = 0u;
}
static void stats_reset(struct stats *s)
{
s->hits = s->misses = 0u;
}
static void stats_level_accessed(struct stats *s, unsigned level)
{
if (level >= s->hit_threshold)
s->hits++;
else
s->misses++;
}
static void stats_miss(struct stats *s)
{
s->misses++;
}
/*
* There are times when we don't have any confidence in the hotspot queue.
* Such as when a fresh cache is created and the blocks have been spread
* out across the levels, or if an io load changes. We detect this by
* seeing how often a lookup is in the top levels of the hotspot queue.
*/
static enum performance stats_assess(struct stats *s)
{
unsigned confidence = safe_div(s->hits << FP_SHIFT, s->hits + s->misses);
if (confidence < SIXTEENTH)
return Q_POOR;
else if (confidence < EIGHTH)
return Q_FAIR;
else
return Q_WELL;
}
/*----------------------------------------------------------------*/
struct smq_hash_table {
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
struct entry_space *es;
unsigned long long hash_bits;
unsigned *buckets;
};
/*
* All cache entries are stored in a chained hash table. To save space we
* use indexing again, and only store indexes to the next entry.
*/
static int h_init(struct smq_hash_table *ht, struct entry_space *es, unsigned nr_entries)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
unsigned i, nr_buckets;
ht->es = es;
nr_buckets = roundup_pow_of_two(max(nr_entries / 4u, 16u));
ht->hash_bits = __ffs(nr_buckets);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
treewide: Use array_size() in vmalloc() The vmalloc() function has no 2-factor argument form, so multiplication factors need to be wrapped in array_size(). This patch replaces cases of: vmalloc(a * b) with: vmalloc(array_size(a, b)) as well as handling cases of: vmalloc(a * b * c) with: vmalloc(array3_size(a, b, c)) This does, however, attempt to ignore constant size factors like: vmalloc(4 * 1024) though any constants defined via macros get caught up in the conversion. Any factors with a sizeof() of "unsigned char", "char", and "u8" were dropped, since they're redundant. The Coccinelle script used for this was: // Fix redundant parens around sizeof(). @@ type TYPE; expression THING, E; @@ ( vmalloc( - (sizeof(TYPE)) * E + sizeof(TYPE) * E , ...) | vmalloc( - (sizeof(THING)) * E + sizeof(THING) * E , ...) ) // Drop single-byte sizes and redundant parens. @@ expression COUNT; typedef u8; typedef __u8; @@ ( vmalloc( - sizeof(u8) * (COUNT) + COUNT , ...) | vmalloc( - sizeof(__u8) * (COUNT) + COUNT , ...) | vmalloc( - sizeof(char) * (COUNT) + COUNT , ...) | vmalloc( - sizeof(unsigned char) * (COUNT) + COUNT , ...) | vmalloc( - sizeof(u8) * COUNT + COUNT , ...) | vmalloc( - sizeof(__u8) * COUNT + COUNT , ...) | vmalloc( - sizeof(char) * COUNT + COUNT , ...) | vmalloc( - sizeof(unsigned char) * COUNT + COUNT , ...) ) // 2-factor product with sizeof(type/expression) and identifier or constant. @@ type TYPE; expression THING; identifier COUNT_ID; constant COUNT_CONST; @@ ( vmalloc( - sizeof(TYPE) * (COUNT_ID) + array_size(COUNT_ID, sizeof(TYPE)) , ...) | vmalloc( - sizeof(TYPE) * COUNT_ID + array_size(COUNT_ID, sizeof(TYPE)) , ...) | vmalloc( - sizeof(TYPE) * (COUNT_CONST) + array_size(COUNT_CONST, sizeof(TYPE)) , ...) | vmalloc( - sizeof(TYPE) * COUNT_CONST + array_size(COUNT_CONST, sizeof(TYPE)) , ...) | vmalloc( - sizeof(THING) * (COUNT_ID) + array_size(COUNT_ID, sizeof(THING)) , ...) | vmalloc( - sizeof(THING) * COUNT_ID + array_size(COUNT_ID, sizeof(THING)) , ...) | vmalloc( - sizeof(THING) * (COUNT_CONST) + array_size(COUNT_CONST, sizeof(THING)) , ...) | vmalloc( - sizeof(THING) * COUNT_CONST + array_size(COUNT_CONST, sizeof(THING)) , ...) ) // 2-factor product, only identifiers. @@ identifier SIZE, COUNT; @@ vmalloc( - SIZE * COUNT + array_size(COUNT, SIZE) , ...) // 3-factor product with 1 sizeof(type) or sizeof(expression), with // redundant parens removed. @@ expression THING; identifier STRIDE, COUNT; type TYPE; @@ ( vmalloc( - sizeof(TYPE) * (COUNT) * (STRIDE) + array3_size(COUNT, STRIDE, sizeof(TYPE)) , ...) | vmalloc( - sizeof(TYPE) * (COUNT) * STRIDE + array3_size(COUNT, STRIDE, sizeof(TYPE)) , ...) | vmalloc( - sizeof(TYPE) * COUNT * (STRIDE) + array3_size(COUNT, STRIDE, sizeof(TYPE)) , ...) | vmalloc( - sizeof(TYPE) * COUNT * STRIDE + array3_size(COUNT, STRIDE, sizeof(TYPE)) , ...) | vmalloc( - sizeof(THING) * (COUNT) * (STRIDE) + array3_size(COUNT, STRIDE, sizeof(THING)) , ...) | vmalloc( - sizeof(THING) * (COUNT) * STRIDE + array3_size(COUNT, STRIDE, sizeof(THING)) , ...) | vmalloc( - sizeof(THING) * COUNT * (STRIDE) + array3_size(COUNT, STRIDE, sizeof(THING)) , ...) | vmalloc( - sizeof(THING) * COUNT * STRIDE + array3_size(COUNT, STRIDE, sizeof(THING)) , ...) ) // 3-factor product with 2 sizeof(variable), with redundant parens removed. @@ expression THING1, THING2; identifier COUNT; type TYPE1, TYPE2; @@ ( vmalloc( - sizeof(TYPE1) * sizeof(TYPE2) * COUNT + array3_size(COUNT, sizeof(TYPE1), sizeof(TYPE2)) , ...) | vmalloc( - sizeof(TYPE1) * sizeof(THING2) * (COUNT) + array3_size(COUNT, sizeof(TYPE1), sizeof(TYPE2)) , ...) | vmalloc( - sizeof(THING1) * sizeof(THING2) * COUNT + array3_size(COUNT, sizeof(THING1), sizeof(THING2)) , ...) | vmalloc( - sizeof(THING1) * sizeof(THING2) * (COUNT) + array3_size(COUNT, sizeof(THING1), sizeof(THING2)) , ...) | vmalloc( - sizeof(TYPE1) * sizeof(THING2) * COUNT + array3_size(COUNT, sizeof(TYPE1), sizeof(THING2)) , ...) | vmalloc( - sizeof(TYPE1) * sizeof(THING2) * (COUNT) + array3_size(COUNT, sizeof(TYPE1), sizeof(THING2)) , ...) ) // 3-factor product, only identifiers, with redundant parens removed. @@ identifier STRIDE, SIZE, COUNT; @@ ( vmalloc( - (COUNT) * STRIDE * SIZE + array3_size(COUNT, STRIDE, SIZE) , ...) | vmalloc( - COUNT * (STRIDE) * SIZE + array3_size(COUNT, STRIDE, SIZE) , ...) | vmalloc( - COUNT * STRIDE * (SIZE) + array3_size(COUNT, STRIDE, SIZE) , ...) | vmalloc( - (COUNT) * (STRIDE) * SIZE + array3_size(COUNT, STRIDE, SIZE) , ...) | vmalloc( - COUNT * (STRIDE) * (SIZE) + array3_size(COUNT, STRIDE, SIZE) , ...) | vmalloc( - (COUNT) * STRIDE * (SIZE) + array3_size(COUNT, STRIDE, SIZE) , ...) | vmalloc( - (COUNT) * (STRIDE) * (SIZE) + array3_size(COUNT, STRIDE, SIZE) , ...) | vmalloc( - COUNT * STRIDE * SIZE + array3_size(COUNT, STRIDE, SIZE) , ...) ) // Any remaining multi-factor products, first at least 3-factor products // when they're not all constants... @@ expression E1, E2, E3; constant C1, C2, C3; @@ ( vmalloc(C1 * C2 * C3, ...) | vmalloc( - E1 * E2 * E3 + array3_size(E1, E2, E3) , ...) ) // And then all remaining 2 factors products when they're not all constants. @@ expression E1, E2; constant C1, C2; @@ ( vmalloc(C1 * C2, ...) | vmalloc( - E1 * E2 + array_size(E1, E2) , ...) ) Signed-off-by: Kees Cook <keescook@chromium.org>
2018-06-13 05:27:11 +08:00
ht->buckets = vmalloc(array_size(nr_buckets, sizeof(*ht->buckets)));
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
if (!ht->buckets)
return -ENOMEM;
for (i = 0; i < nr_buckets; i++)
ht->buckets[i] = INDEXER_NULL;
return 0;
}
static void h_exit(struct smq_hash_table *ht)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
vfree(ht->buckets);
}
static struct entry *h_head(struct smq_hash_table *ht, unsigned bucket)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
return to_entry(ht->es, ht->buckets[bucket]);
}
static struct entry *h_next(struct smq_hash_table *ht, struct entry *e)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
return to_entry(ht->es, e->hash_next);
}
static void __h_insert(struct smq_hash_table *ht, unsigned bucket, struct entry *e)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
e->hash_next = ht->buckets[bucket];
ht->buckets[bucket] = to_index(ht->es, e);
}
static void h_insert(struct smq_hash_table *ht, struct entry *e)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
unsigned h = hash_64(from_oblock(e->oblock), ht->hash_bits);
__h_insert(ht, h, e);
}
static struct entry *__h_lookup(struct smq_hash_table *ht, unsigned h, dm_oblock_t oblock,
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
struct entry **prev)
{
struct entry *e;
*prev = NULL;
for (e = h_head(ht, h); e; e = h_next(ht, e)) {
if (e->oblock == oblock)
return e;
*prev = e;
}
return NULL;
}
static void __h_unlink(struct smq_hash_table *ht, unsigned h,
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
struct entry *e, struct entry *prev)
{
if (prev)
prev->hash_next = e->hash_next;
else
ht->buckets[h] = e->hash_next;
}
/*
* Also moves each entry to the front of the bucket.
*/
static struct entry *h_lookup(struct smq_hash_table *ht, dm_oblock_t oblock)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
struct entry *e, *prev;
unsigned h = hash_64(from_oblock(oblock), ht->hash_bits);
e = __h_lookup(ht, h, oblock, &prev);
if (e && prev) {
/*
* Move to the front because this entry is likely
* to be hit again.
*/
__h_unlink(ht, h, e, prev);
__h_insert(ht, h, e);
}
return e;
}
static void h_remove(struct smq_hash_table *ht, struct entry *e)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
unsigned h = hash_64(from_oblock(e->oblock), ht->hash_bits);
struct entry *prev;
/*
* The down side of using a singly linked list is we have to
* iterate the bucket to remove an item.
*/
e = __h_lookup(ht, h, e->oblock, &prev);
if (e)
__h_unlink(ht, h, e, prev);
}
/*----------------------------------------------------------------*/
struct entry_alloc {
struct entry_space *es;
unsigned begin;
unsigned nr_allocated;
struct ilist free;
};
static void init_allocator(struct entry_alloc *ea, struct entry_space *es,
unsigned begin, unsigned end)
{
unsigned i;
ea->es = es;
ea->nr_allocated = 0u;
ea->begin = begin;
l_init(&ea->free);
for (i = begin; i != end; i++)
l_add_tail(ea->es, &ea->free, __get_entry(ea->es, i));
}
static void init_entry(struct entry *e)
{
/*
* We can't memset because that would clear the hotspot and
* sentinel bits which remain constant.
*/
e->hash_next = INDEXER_NULL;
e->next = INDEXER_NULL;
e->prev = INDEXER_NULL;
e->level = 0u;
e->dirty = true; /* FIXME: audit */
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
e->allocated = true;
e->sentinel = false;
e->pending_work = false;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
}
static struct entry *alloc_entry(struct entry_alloc *ea)
{
struct entry *e;
if (l_empty(&ea->free))
return NULL;
e = l_pop_head(ea->es, &ea->free);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
init_entry(e);
ea->nr_allocated++;
return e;
}
/*
* This assumes the cblock hasn't already been allocated.
*/
static struct entry *alloc_particular_entry(struct entry_alloc *ea, unsigned i)
{
struct entry *e = __get_entry(ea->es, ea->begin + i);
BUG_ON(e->allocated);
l_del(ea->es, &ea->free, e);
init_entry(e);
ea->nr_allocated++;
return e;
}
static void free_entry(struct entry_alloc *ea, struct entry *e)
{
BUG_ON(!ea->nr_allocated);
BUG_ON(!e->allocated);
ea->nr_allocated--;
e->allocated = false;
l_add_tail(ea->es, &ea->free, e);
}
static bool allocator_empty(struct entry_alloc *ea)
{
return l_empty(&ea->free);
}
static unsigned get_index(struct entry_alloc *ea, struct entry *e)
{
return to_index(ea->es, e) - ea->begin;
}
static struct entry *get_entry(struct entry_alloc *ea, unsigned index)
{
return __get_entry(ea->es, ea->begin + index);
}
/*----------------------------------------------------------------*/
#define NR_HOTSPOT_LEVELS 64u
#define NR_CACHE_LEVELS 64u
#define WRITEBACK_PERIOD (10ul * HZ)
#define DEMOTE_PERIOD (60ul * HZ)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
#define HOTSPOT_UPDATE_PERIOD (HZ)
#define CACHE_UPDATE_PERIOD (60ul * HZ)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
struct smq_policy {
struct dm_cache_policy policy;
/* protects everything */
spinlock_t lock;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
dm_cblock_t cache_size;
sector_t cache_block_size;
sector_t hotspot_block_size;
unsigned nr_hotspot_blocks;
unsigned cache_blocks_per_hotspot_block;
unsigned hotspot_level_jump;
struct entry_space es;
struct entry_alloc writeback_sentinel_alloc;
struct entry_alloc demote_sentinel_alloc;
struct entry_alloc hotspot_alloc;
struct entry_alloc cache_alloc;
unsigned long *hotspot_hit_bits;
unsigned long *cache_hit_bits;
/*
* We maintain three queues of entries. The cache proper,
* consisting of a clean and dirty queue, containing the currently
* active mappings. The hotspot queue uses a larger block size to
* track blocks that are being hit frequently and potential
* candidates for promotion to the cache.
*/
struct queue hotspot;
struct queue clean;
struct queue dirty;
struct stats hotspot_stats;
struct stats cache_stats;
/*
* Keeps track of time, incremented by the core. We use this to
* avoid attributing multiple hits within the same tick.
*/
unsigned tick;
/*
* The hash tables allows us to quickly find an entry by origin
* block.
*/
struct smq_hash_table table;
struct smq_hash_table hotspot_table;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
bool current_writeback_sentinels;
unsigned long next_writeback_period;
bool current_demote_sentinels;
unsigned long next_demote_period;
unsigned write_promote_level;
unsigned read_promote_level;
unsigned long next_hotspot_period;
unsigned long next_cache_period;
struct background_tracker *bg_work;
bool migrations_allowed;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
};
/*----------------------------------------------------------------*/
static struct entry *get_sentinel(struct entry_alloc *ea, unsigned level, bool which)
{
return get_entry(ea, which ? level : NR_CACHE_LEVELS + level);
}
static struct entry *writeback_sentinel(struct smq_policy *mq, unsigned level)
{
return get_sentinel(&mq->writeback_sentinel_alloc, level, mq->current_writeback_sentinels);
}
static struct entry *demote_sentinel(struct smq_policy *mq, unsigned level)
{
return get_sentinel(&mq->demote_sentinel_alloc, level, mq->current_demote_sentinels);
}
static void __update_writeback_sentinels(struct smq_policy *mq)
{
unsigned level;
struct queue *q = &mq->dirty;
struct entry *sentinel;
for (level = 0; level < q->nr_levels; level++) {
sentinel = writeback_sentinel(mq, level);
q_del(q, sentinel);
q_push(q, sentinel);
}
}
static void __update_demote_sentinels(struct smq_policy *mq)
{
unsigned level;
struct queue *q = &mq->clean;
struct entry *sentinel;
for (level = 0; level < q->nr_levels; level++) {
sentinel = demote_sentinel(mq, level);
q_del(q, sentinel);
q_push(q, sentinel);
}
}
static void update_sentinels(struct smq_policy *mq)
{
if (time_after(jiffies, mq->next_writeback_period)) {
mq->next_writeback_period = jiffies + WRITEBACK_PERIOD;
mq->current_writeback_sentinels = !mq->current_writeback_sentinels;
__update_writeback_sentinels(mq);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
}
if (time_after(jiffies, mq->next_demote_period)) {
mq->next_demote_period = jiffies + DEMOTE_PERIOD;
mq->current_demote_sentinels = !mq->current_demote_sentinels;
__update_demote_sentinels(mq);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
}
}
static void __sentinels_init(struct smq_policy *mq)
{
unsigned level;
struct entry *sentinel;
for (level = 0; level < NR_CACHE_LEVELS; level++) {
sentinel = writeback_sentinel(mq, level);
sentinel->level = level;
q_push(&mq->dirty, sentinel);
sentinel = demote_sentinel(mq, level);
sentinel->level = level;
q_push(&mq->clean, sentinel);
}
}
static void sentinels_init(struct smq_policy *mq)
{
mq->next_writeback_period = jiffies + WRITEBACK_PERIOD;
mq->next_demote_period = jiffies + DEMOTE_PERIOD;
mq->current_writeback_sentinels = false;
mq->current_demote_sentinels = false;
__sentinels_init(mq);
mq->current_writeback_sentinels = !mq->current_writeback_sentinels;
mq->current_demote_sentinels = !mq->current_demote_sentinels;
__sentinels_init(mq);
}
/*----------------------------------------------------------------*/
static void del_queue(struct smq_policy *mq, struct entry *e)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
q_del(e->dirty ? &mq->dirty : &mq->clean, e);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
}
static void push_queue(struct smq_policy *mq, struct entry *e)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
if (e->dirty)
q_push(&mq->dirty, e);
else
q_push(&mq->clean, e);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
}
// !h, !q, a -> h, q, a
static void push(struct smq_policy *mq, struct entry *e)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
h_insert(&mq->table, e);
if (!e->pending_work)
push_queue(mq, e);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
}
static void push_queue_front(struct smq_policy *mq, struct entry *e)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
if (e->dirty)
q_push_front(&mq->dirty, e);
else
q_push_front(&mq->clean, e);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
}
static void push_front(struct smq_policy *mq, struct entry *e)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
h_insert(&mq->table, e);
if (!e->pending_work)
push_queue_front(mq, e);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
}
static dm_cblock_t infer_cblock(struct smq_policy *mq, struct entry *e)
{
return to_cblock(get_index(&mq->cache_alloc, e));
}
static void requeue(struct smq_policy *mq, struct entry *e)
{
/*
* Pending work has temporarily been taken out of the queues.
*/
if (e->pending_work)
return;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
if (!test_and_set_bit(from_cblock(infer_cblock(mq, e)), mq->cache_hit_bits)) {
if (!e->dirty) {
q_requeue(&mq->clean, e, 1u, NULL, NULL);
return;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
}
q_requeue(&mq->dirty, e, 1u,
get_sentinel(&mq->writeback_sentinel_alloc, e->level, !mq->current_writeback_sentinels),
get_sentinel(&mq->writeback_sentinel_alloc, e->level, mq->current_writeback_sentinels));
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
}
}
static unsigned default_promote_level(struct smq_policy *mq)
{
/*
* The promote level depends on the current performance of the
* cache.
*
* If the cache is performing badly, then we can't afford
* to promote much without causing performance to drop below that
* of the origin device.
*
* If the cache is performing well, then we don't need to promote
* much. If it isn't broken, don't fix it.
*
* If the cache is middling then we promote more.
*
* This scheme reminds me of a graph of entropy vs probability of a
* binary variable.
*/
static unsigned table[] = {1, 1, 1, 2, 4, 6, 7, 8, 7, 6, 4, 4, 3, 3, 2, 2, 1};
unsigned hits = mq->cache_stats.hits;
unsigned misses = mq->cache_stats.misses;
unsigned index = safe_div(hits << 4u, hits + misses);
return table[index];
}
static void update_promote_levels(struct smq_policy *mq)
{
/*
* If there are unused cache entries then we want to be really
* eager to promote.
*/
unsigned threshold_level = allocator_empty(&mq->cache_alloc) ?
default_promote_level(mq) : (NR_HOTSPOT_LEVELS / 2u);
threshold_level = max(threshold_level, NR_HOTSPOT_LEVELS);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
/*
* If the hotspot queue is performing badly then we have little
* confidence that we know which blocks to promote. So we cut down
* the amount of promotions.
*/
switch (stats_assess(&mq->hotspot_stats)) {
case Q_POOR:
threshold_level /= 4u;
break;
case Q_FAIR:
threshold_level /= 2u;
break;
case Q_WELL:
break;
}
mq->read_promote_level = NR_HOTSPOT_LEVELS - threshold_level;
mq->write_promote_level = (NR_HOTSPOT_LEVELS - threshold_level);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
}
/*
* If the hotspot queue is performing badly, then we try and move entries
* around more quickly.
*/
static void update_level_jump(struct smq_policy *mq)
{
switch (stats_assess(&mq->hotspot_stats)) {
case Q_POOR:
mq->hotspot_level_jump = 4u;
break;
case Q_FAIR:
mq->hotspot_level_jump = 2u;
break;
case Q_WELL:
mq->hotspot_level_jump = 1u;
break;
}
}
static void end_hotspot_period(struct smq_policy *mq)
{
clear_bitset(mq->hotspot_hit_bits, mq->nr_hotspot_blocks);
update_promote_levels(mq);
if (time_after(jiffies, mq->next_hotspot_period)) {
update_level_jump(mq);
q_redistribute(&mq->hotspot);
stats_reset(&mq->hotspot_stats);
mq->next_hotspot_period = jiffies + HOTSPOT_UPDATE_PERIOD;
}
}
static void end_cache_period(struct smq_policy *mq)
{
if (time_after(jiffies, mq->next_cache_period)) {
clear_bitset(mq->cache_hit_bits, from_cblock(mq->cache_size));
q_redistribute(&mq->dirty);
q_redistribute(&mq->clean);
stats_reset(&mq->cache_stats);
mq->next_cache_period = jiffies + CACHE_UPDATE_PERIOD;
}
}
/*----------------------------------------------------------------*/
/*
* Targets are given as a percentage.
*/
#define CLEAN_TARGET 25u
#define FREE_TARGET 25u
static unsigned percent_to_target(struct smq_policy *mq, unsigned p)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
return from_cblock(mq->cache_size) * p / 100u;
}
static bool clean_target_met(struct smq_policy *mq, bool idle)
{
/*
* Cache entries may not be populated. So we cannot rely on the
* size of the clean queue.
*/
if (idle) {
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
/*
* We'd like to clean everything.
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
*/
return q_size(&mq->dirty) == 0u;
}
/*
* If we're busy we don't worry about cleaning at all.
*/
return true;
}
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
static bool free_target_met(struct smq_policy *mq)
{
unsigned nr_free;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
nr_free = from_cblock(mq->cache_size) - mq->cache_alloc.nr_allocated;
return (nr_free + btracker_nr_demotions_queued(mq->bg_work)) >=
percent_to_target(mq, FREE_TARGET);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
}
/*----------------------------------------------------------------*/
static void mark_pending(struct smq_policy *mq, struct entry *e)
{
BUG_ON(e->sentinel);
BUG_ON(!e->allocated);
BUG_ON(e->pending_work);
e->pending_work = true;
}
static void clear_pending(struct smq_policy *mq, struct entry *e)
{
BUG_ON(!e->pending_work);
e->pending_work = false;
}
static void queue_writeback(struct smq_policy *mq, bool idle)
{
int r;
struct policy_work work;
struct entry *e;
e = q_peek(&mq->dirty, mq->dirty.nr_levels, idle);
if (e) {
mark_pending(mq, e);
q_del(&mq->dirty, e);
work.op = POLICY_WRITEBACK;
work.oblock = e->oblock;
work.cblock = infer_cblock(mq, e);
r = btracker_queue(mq->bg_work, &work, NULL);
if (r) {
clear_pending(mq, e);
q_push_front(&mq->dirty, e);
}
}
}
static void queue_demotion(struct smq_policy *mq)
{
int r;
struct policy_work work;
struct entry *e;
if (WARN_ON_ONCE(!mq->migrations_allowed))
return;
e = q_peek(&mq->clean, mq->clean.nr_levels / 2, true);
if (!e) {
if (!clean_target_met(mq, true))
queue_writeback(mq, false);
return;
}
mark_pending(mq, e);
q_del(&mq->clean, e);
work.op = POLICY_DEMOTE;
work.oblock = e->oblock;
work.cblock = infer_cblock(mq, e);
r = btracker_queue(mq->bg_work, &work, NULL);
if (r) {
clear_pending(mq, e);
q_push_front(&mq->clean, e);
}
}
static void queue_promotion(struct smq_policy *mq, dm_oblock_t oblock,
struct policy_work **workp)
{
int r;
struct entry *e;
struct policy_work work;
if (!mq->migrations_allowed)
return;
if (allocator_empty(&mq->cache_alloc)) {
/*
* We always claim to be 'idle' to ensure some demotions happen
* with continuous loads.
*/
if (!free_target_met(mq))
queue_demotion(mq);
return;
}
if (btracker_promotion_already_present(mq->bg_work, oblock))
return;
/*
* We allocate the entry now to reserve the cblock. If the
* background work is aborted we must remember to free it.
*/
e = alloc_entry(&mq->cache_alloc);
BUG_ON(!e);
e->pending_work = true;
work.op = POLICY_PROMOTE;
work.oblock = oblock;
work.cblock = infer_cblock(mq, e);
r = btracker_queue(mq->bg_work, &work, workp);
if (r)
free_entry(&mq->cache_alloc, e);
}
/*----------------------------------------------------------------*/
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
enum promote_result {
PROMOTE_NOT,
PROMOTE_TEMPORARY,
PROMOTE_PERMANENT
};
/*
* Converts a boolean into a promote result.
*/
static enum promote_result maybe_promote(bool promote)
{
return promote ? PROMOTE_PERMANENT : PROMOTE_NOT;
}
static enum promote_result should_promote(struct smq_policy *mq, struct entry *hs_e,
int data_dir, bool fast_promote)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
if (data_dir == WRITE) {
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
if (!allocator_empty(&mq->cache_alloc) && fast_promote)
return PROMOTE_TEMPORARY;
return maybe_promote(hs_e->level >= mq->write_promote_level);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
} else
return maybe_promote(hs_e->level >= mq->read_promote_level);
}
static dm_oblock_t to_hblock(struct smq_policy *mq, dm_oblock_t b)
{
sector_t r = from_oblock(b);
(void) sector_div(r, mq->cache_blocks_per_hotspot_block);
return to_oblock(r);
}
static struct entry *update_hotspot_queue(struct smq_policy *mq, dm_oblock_t b)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
unsigned hi;
dm_oblock_t hb = to_hblock(mq, b);
struct entry *e = h_lookup(&mq->hotspot_table, hb);
if (e) {
stats_level_accessed(&mq->hotspot_stats, e->level);
hi = get_index(&mq->hotspot_alloc, e);
q_requeue(&mq->hotspot, e,
test_and_set_bit(hi, mq->hotspot_hit_bits) ?
0u : mq->hotspot_level_jump,
NULL, NULL);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
} else {
stats_miss(&mq->hotspot_stats);
e = alloc_entry(&mq->hotspot_alloc);
if (!e) {
e = q_pop(&mq->hotspot);
if (e) {
h_remove(&mq->hotspot_table, e);
hi = get_index(&mq->hotspot_alloc, e);
clear_bit(hi, mq->hotspot_hit_bits);
}
}
if (e) {
e->oblock = hb;
q_push(&mq->hotspot, e);
h_insert(&mq->hotspot_table, e);
}
}
return e;
}
/*----------------------------------------------------------------*/
/*
* Public interface, via the policy struct. See dm-cache-policy.h for a
* description of these.
*/
static struct smq_policy *to_smq_policy(struct dm_cache_policy *p)
{
return container_of(p, struct smq_policy, policy);
}
static void smq_destroy(struct dm_cache_policy *p)
{
struct smq_policy *mq = to_smq_policy(p);
btracker_destroy(mq->bg_work);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
h_exit(&mq->hotspot_table);
h_exit(&mq->table);
free_bitset(mq->hotspot_hit_bits);
free_bitset(mq->cache_hit_bits);
space_exit(&mq->es);
kfree(mq);
}
/*----------------------------------------------------------------*/
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
static int __lookup(struct smq_policy *mq, dm_oblock_t oblock, dm_cblock_t *cblock,
int data_dir, bool fast_copy,
struct policy_work **work, bool *background_work)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
struct entry *e, *hs_e;
enum promote_result pr;
*background_work = false;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
e = h_lookup(&mq->table, oblock);
if (e) {
stats_level_accessed(&mq->cache_stats, e->level);
requeue(mq, e);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
*cblock = infer_cblock(mq, e);
return 0;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
} else {
stats_miss(&mq->cache_stats);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
/*
* The hotspot queue only gets updated with misses.
*/
hs_e = update_hotspot_queue(mq, oblock);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
pr = should_promote(mq, hs_e, data_dir, fast_copy);
if (pr != PROMOTE_NOT) {
queue_promotion(mq, oblock, work);
*background_work = true;
}
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
return -ENOENT;
}
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
}
static int smq_lookup(struct dm_cache_policy *p, dm_oblock_t oblock, dm_cblock_t *cblock,
int data_dir, bool fast_copy,
bool *background_work)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
int r;
unsigned long flags;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
struct smq_policy *mq = to_smq_policy(p);
spin_lock_irqsave(&mq->lock, flags);
r = __lookup(mq, oblock, cblock,
data_dir, fast_copy,
NULL, background_work);
spin_unlock_irqrestore(&mq->lock, flags);
return r;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
}
static int smq_lookup_with_work(struct dm_cache_policy *p,
dm_oblock_t oblock, dm_cblock_t *cblock,
int data_dir, bool fast_copy,
struct policy_work **work)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
int r;
bool background_queued;
unsigned long flags;
struct smq_policy *mq = to_smq_policy(p);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
spin_lock_irqsave(&mq->lock, flags);
r = __lookup(mq, oblock, cblock, data_dir, fast_copy, work, &background_queued);
spin_unlock_irqrestore(&mq->lock, flags);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
return r;
}
static int smq_get_background_work(struct dm_cache_policy *p, bool idle,
struct policy_work **result)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
int r;
unsigned long flags;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
struct smq_policy *mq = to_smq_policy(p);
spin_lock_irqsave(&mq->lock, flags);
r = btracker_issue(mq->bg_work, result);
if (r == -ENODATA) {
if (!clean_target_met(mq, idle)) {
queue_writeback(mq, idle);
r = btracker_issue(mq->bg_work, result);
}
}
spin_unlock_irqrestore(&mq->lock, flags);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
return r;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
}
/*
* We need to clear any pending work flags that have been set, and in the
* case of promotion free the entry for the destination cblock.
*/
static void __complete_background_work(struct smq_policy *mq,
struct policy_work *work,
bool success)
{
struct entry *e = get_entry(&mq->cache_alloc,
from_cblock(work->cblock));
switch (work->op) {
case POLICY_PROMOTE:
// !h, !q, a
clear_pending(mq, e);
if (success) {
e->oblock = work->oblock;
e->level = NR_CACHE_LEVELS - 1;
push(mq, e);
// h, q, a
} else {
free_entry(&mq->cache_alloc, e);
// !h, !q, !a
}
break;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
case POLICY_DEMOTE:
// h, !q, a
if (success) {
h_remove(&mq->table, e);
free_entry(&mq->cache_alloc, e);
// !h, !q, !a
} else {
clear_pending(mq, e);
push_queue(mq, e);
// h, q, a
}
break;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
case POLICY_WRITEBACK:
// h, !q, a
clear_pending(mq, e);
push_queue(mq, e);
// h, q, a
break;
}
btracker_complete(mq->bg_work, work);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
}
static void smq_complete_background_work(struct dm_cache_policy *p,
struct policy_work *work,
bool success)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
unsigned long flags;
struct smq_policy *mq = to_smq_policy(p);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
spin_lock_irqsave(&mq->lock, flags);
__complete_background_work(mq, work, success);
spin_unlock_irqrestore(&mq->lock, flags);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
}
// in_hash(oblock) -> in_hash(oblock)
static void __smq_set_clear_dirty(struct smq_policy *mq, dm_cblock_t cblock, bool set)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
struct entry *e = get_entry(&mq->cache_alloc, from_cblock(cblock));
if (e->pending_work)
e->dirty = set;
else {
del_queue(mq, e);
e->dirty = set;
push_queue(mq, e);
}
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
}
static void smq_set_dirty(struct dm_cache_policy *p, dm_cblock_t cblock)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
unsigned long flags;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
struct smq_policy *mq = to_smq_policy(p);
spin_lock_irqsave(&mq->lock, flags);
__smq_set_clear_dirty(mq, cblock, true);
spin_unlock_irqrestore(&mq->lock, flags);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
}
static void smq_clear_dirty(struct dm_cache_policy *p, dm_cblock_t cblock)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
struct smq_policy *mq = to_smq_policy(p);
unsigned long flags;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
spin_lock_irqsave(&mq->lock, flags);
__smq_set_clear_dirty(mq, cblock, false);
spin_unlock_irqrestore(&mq->lock, flags);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
}
static unsigned random_level(dm_cblock_t cblock)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
return hash_32(from_cblock(cblock), 9) & (NR_CACHE_LEVELS - 1);
}
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
static int smq_load_mapping(struct dm_cache_policy *p,
dm_oblock_t oblock, dm_cblock_t cblock,
bool dirty, uint32_t hint, bool hint_valid)
{
struct smq_policy *mq = to_smq_policy(p);
struct entry *e;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
e = alloc_particular_entry(&mq->cache_alloc, from_cblock(cblock));
e->oblock = oblock;
e->dirty = dirty;
e->level = hint_valid ? min(hint, NR_CACHE_LEVELS - 1) : random_level(cblock);
e->pending_work = false;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
/*
* When we load mappings we push ahead of both sentinels in order to
* allow demotions and cleaning to occur immediately.
*/
push_front(mq, e);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
return 0;
}
static int smq_invalidate_mapping(struct dm_cache_policy *p, dm_cblock_t cblock)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
struct smq_policy *mq = to_smq_policy(p);
struct entry *e = get_entry(&mq->cache_alloc, from_cblock(cblock));
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
if (!e->allocated)
return -ENODATA;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
// FIXME: what if this block has pending background work?
del_queue(mq, e);
h_remove(&mq->table, e);
free_entry(&mq->cache_alloc, e);
return 0;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
}
static uint32_t smq_get_hint(struct dm_cache_policy *p, dm_cblock_t cblock)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
struct smq_policy *mq = to_smq_policy(p);
struct entry *e = get_entry(&mq->cache_alloc, from_cblock(cblock));
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
if (!e->allocated)
return 0;
return e->level;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
}
static dm_cblock_t smq_residency(struct dm_cache_policy *p)
{
dm_cblock_t r;
unsigned long flags;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
struct smq_policy *mq = to_smq_policy(p);
spin_lock_irqsave(&mq->lock, flags);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
r = to_cblock(mq->cache_alloc.nr_allocated);
spin_unlock_irqrestore(&mq->lock, flags);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
return r;
}
static void smq_tick(struct dm_cache_policy *p, bool can_block)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
struct smq_policy *mq = to_smq_policy(p);
unsigned long flags;
spin_lock_irqsave(&mq->lock, flags);
mq->tick++;
update_sentinels(mq);
end_hotspot_period(mq);
end_cache_period(mq);
spin_unlock_irqrestore(&mq->lock, flags);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
}
static void smq_allow_migrations(struct dm_cache_policy *p, bool allow)
{
struct smq_policy *mq = to_smq_policy(p);
mq->migrations_allowed = allow;
}
/*
* smq has no config values, but the old mq policy did. To avoid breaking
* software we continue to accept these configurables for the mq policy,
* but they have no effect.
*/
static int mq_set_config_value(struct dm_cache_policy *p,
const char *key, const char *value)
{
unsigned long tmp;
if (kstrtoul(value, 10, &tmp))
return -EINVAL;
if (!strcasecmp(key, "random_threshold") ||
!strcasecmp(key, "sequential_threshold") ||
!strcasecmp(key, "discard_promote_adjustment") ||
!strcasecmp(key, "read_promote_adjustment") ||
!strcasecmp(key, "write_promote_adjustment")) {
DMWARN("tunable '%s' no longer has any effect, mq policy is now an alias for smq", key);
return 0;
}
return -EINVAL;
}
static int mq_emit_config_values(struct dm_cache_policy *p, char *result,
unsigned maxlen, ssize_t *sz_ptr)
{
ssize_t sz = *sz_ptr;
DMEMIT("10 random_threshold 0 "
"sequential_threshold 0 "
"discard_promote_adjustment 0 "
"read_promote_adjustment 0 "
"write_promote_adjustment 0 ");
*sz_ptr = sz;
return 0;
}
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
/* Init the policy plugin interface function pointers. */
static void init_policy_functions(struct smq_policy *mq, bool mimic_mq)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
mq->policy.destroy = smq_destroy;
mq->policy.lookup = smq_lookup;
mq->policy.lookup_with_work = smq_lookup_with_work;
mq->policy.get_background_work = smq_get_background_work;
mq->policy.complete_background_work = smq_complete_background_work;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
mq->policy.set_dirty = smq_set_dirty;
mq->policy.clear_dirty = smq_clear_dirty;
mq->policy.load_mapping = smq_load_mapping;
mq->policy.invalidate_mapping = smq_invalidate_mapping;
mq->policy.get_hint = smq_get_hint;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
mq->policy.residency = smq_residency;
mq->policy.tick = smq_tick;
mq->policy.allow_migrations = smq_allow_migrations;
if (mimic_mq) {
mq->policy.set_config_value = mq_set_config_value;
mq->policy.emit_config_values = mq_emit_config_values;
}
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
}
static bool too_many_hotspot_blocks(sector_t origin_size,
sector_t hotspot_block_size,
unsigned nr_hotspot_blocks)
{
return (hotspot_block_size * nr_hotspot_blocks) > origin_size;
}
static void calc_hotspot_params(sector_t origin_size,
sector_t cache_block_size,
unsigned nr_cache_blocks,
sector_t *hotspot_block_size,
unsigned *nr_hotspot_blocks)
{
*hotspot_block_size = cache_block_size * 16u;
*nr_hotspot_blocks = max(nr_cache_blocks / 4u, 1024u);
while ((*hotspot_block_size > cache_block_size) &&
too_many_hotspot_blocks(origin_size, *hotspot_block_size, *nr_hotspot_blocks))
*hotspot_block_size /= 2u;
}
static struct dm_cache_policy *__smq_create(dm_cblock_t cache_size,
sector_t origin_size,
sector_t cache_block_size,
bool mimic_mq,
bool migrations_allowed)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
{
unsigned i;
unsigned nr_sentinels_per_queue = 2u * NR_CACHE_LEVELS;
unsigned total_sentinels = 2u * nr_sentinels_per_queue;
struct smq_policy *mq = kzalloc(sizeof(*mq), GFP_KERNEL);
if (!mq)
return NULL;
init_policy_functions(mq, mimic_mq);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
mq->cache_size = cache_size;
mq->cache_block_size = cache_block_size;
calc_hotspot_params(origin_size, cache_block_size, from_cblock(cache_size),
&mq->hotspot_block_size, &mq->nr_hotspot_blocks);
mq->cache_blocks_per_hotspot_block = div64_u64(mq->hotspot_block_size, mq->cache_block_size);
mq->hotspot_level_jump = 1u;
if (space_init(&mq->es, total_sentinels + mq->nr_hotspot_blocks + from_cblock(cache_size))) {
DMERR("couldn't initialize entry space");
goto bad_pool_init;
}
init_allocator(&mq->writeback_sentinel_alloc, &mq->es, 0, nr_sentinels_per_queue);
for (i = 0; i < nr_sentinels_per_queue; i++)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
get_entry(&mq->writeback_sentinel_alloc, i)->sentinel = true;
init_allocator(&mq->demote_sentinel_alloc, &mq->es, nr_sentinels_per_queue, total_sentinels);
for (i = 0; i < nr_sentinels_per_queue; i++)
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
get_entry(&mq->demote_sentinel_alloc, i)->sentinel = true;
init_allocator(&mq->hotspot_alloc, &mq->es, total_sentinels,
total_sentinels + mq->nr_hotspot_blocks);
init_allocator(&mq->cache_alloc, &mq->es,
total_sentinels + mq->nr_hotspot_blocks,
total_sentinels + mq->nr_hotspot_blocks + from_cblock(cache_size));
mq->hotspot_hit_bits = alloc_bitset(mq->nr_hotspot_blocks);
if (!mq->hotspot_hit_bits) {
DMERR("couldn't allocate hotspot hit bitset");
goto bad_hotspot_hit_bits;
}
clear_bitset(mq->hotspot_hit_bits, mq->nr_hotspot_blocks);
if (from_cblock(cache_size)) {
mq->cache_hit_bits = alloc_bitset(from_cblock(cache_size));
if (!mq->cache_hit_bits) {
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
DMERR("couldn't allocate cache hit bitset");
goto bad_cache_hit_bits;
}
clear_bitset(mq->cache_hit_bits, from_cblock(mq->cache_size));
} else
mq->cache_hit_bits = NULL;
mq->tick = 0;
spin_lock_init(&mq->lock);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
q_init(&mq->hotspot, &mq->es, NR_HOTSPOT_LEVELS);
mq->hotspot.nr_top_levels = 8;
mq->hotspot.nr_in_top_levels = min(mq->nr_hotspot_blocks / NR_HOTSPOT_LEVELS,
from_cblock(mq->cache_size) / mq->cache_blocks_per_hotspot_block);
q_init(&mq->clean, &mq->es, NR_CACHE_LEVELS);
q_init(&mq->dirty, &mq->es, NR_CACHE_LEVELS);
stats_init(&mq->hotspot_stats, NR_HOTSPOT_LEVELS);
stats_init(&mq->cache_stats, NR_CACHE_LEVELS);
if (h_init(&mq->table, &mq->es, from_cblock(cache_size)))
goto bad_alloc_table;
if (h_init(&mq->hotspot_table, &mq->es, mq->nr_hotspot_blocks))
goto bad_alloc_hotspot_table;
sentinels_init(mq);
mq->write_promote_level = mq->read_promote_level = NR_HOTSPOT_LEVELS;
mq->next_hotspot_period = jiffies;
mq->next_cache_period = jiffies;
mq->bg_work = btracker_create(4096); /* FIXME: hard coded value */
if (!mq->bg_work)
goto bad_btracker;
mq->migrations_allowed = migrations_allowed;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
return &mq->policy;
bad_btracker:
h_exit(&mq->hotspot_table);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
bad_alloc_hotspot_table:
h_exit(&mq->table);
bad_alloc_table:
free_bitset(mq->cache_hit_bits);
bad_cache_hit_bits:
free_bitset(mq->hotspot_hit_bits);
bad_hotspot_hit_bits:
space_exit(&mq->es);
bad_pool_init:
kfree(mq);
return NULL;
}
static struct dm_cache_policy *smq_create(dm_cblock_t cache_size,
sector_t origin_size,
sector_t cache_block_size)
{
return __smq_create(cache_size, origin_size, cache_block_size, false, true);
}
static struct dm_cache_policy *mq_create(dm_cblock_t cache_size,
sector_t origin_size,
sector_t cache_block_size)
{
return __smq_create(cache_size, origin_size, cache_block_size, true, true);
}
static struct dm_cache_policy *cleaner_create(dm_cblock_t cache_size,
sector_t origin_size,
sector_t cache_block_size)
{
return __smq_create(cache_size, origin_size, cache_block_size, false, false);
}
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
/*----------------------------------------------------------------*/
static struct dm_cache_policy_type smq_policy_type = {
.name = "smq",
.version = {2, 0, 0},
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
.hint_size = 4,
.owner = THIS_MODULE,
.create = smq_create
};
static struct dm_cache_policy_type mq_policy_type = {
.name = "mq",
.version = {2, 0, 0},
.hint_size = 4,
.owner = THIS_MODULE,
.create = mq_create,
};
static struct dm_cache_policy_type cleaner_policy_type = {
.name = "cleaner",
.version = {2, 0, 0},
.hint_size = 4,
.owner = THIS_MODULE,
.create = cleaner_create,
};
static struct dm_cache_policy_type default_policy_type = {
.name = "default",
.version = {2, 0, 0},
.hint_size = 4,
.owner = THIS_MODULE,
.create = smq_create,
.real = &smq_policy_type
};
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
static int __init smq_init(void)
{
int r;
r = dm_cache_policy_register(&smq_policy_type);
if (r) {
DMERR("register failed %d", r);
return -ENOMEM;
}
r = dm_cache_policy_register(&mq_policy_type);
if (r) {
DMERR("register failed (as mq) %d", r);
goto out_mq;
}
r = dm_cache_policy_register(&cleaner_policy_type);
if (r) {
DMERR("register failed (as cleaner) %d", r);
goto out_cleaner;
}
r = dm_cache_policy_register(&default_policy_type);
if (r) {
DMERR("register failed (as default) %d", r);
goto out_default;
}
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
return 0;
out_default:
dm_cache_policy_unregister(&cleaner_policy_type);
out_cleaner:
dm_cache_policy_unregister(&mq_policy_type);
out_mq:
dm_cache_policy_unregister(&smq_policy_type);
return -ENOMEM;
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
}
static void __exit smq_exit(void)
{
dm_cache_policy_unregister(&cleaner_policy_type);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
dm_cache_policy_unregister(&smq_policy_type);
dm_cache_policy_unregister(&mq_policy_type);
dm_cache_policy_unregister(&default_policy_type);
dm cache: add stochastic-multi-queue (smq) policy The stochastic-multi-queue (smq) policy addresses some of the problems with the current multiqueue (mq) policy. Memory usage ------------ The mq policy uses a lot of memory; 88 bytes per cache block on a 64 bit machine. SMQ uses 28bit indexes to implement it's data structures rather than pointers. It avoids storing an explicit hit count for each block. It has a 'hotspot' queue rather than a pre cache which uses a quarter of the entries (each hotspot block covers a larger area than a single cache block). All these mean smq uses ~25bytes per cache block. Still a lot of memory, but a substantial improvement nontheless. Level balancing --------------- MQ places entries in different levels of the multiqueue structures based on their hit count (~ln(hit count)). This means the bottom levels generally have the most entries, and the top ones have very few. Having unbalanced levels like this reduces the efficacy of the multiqueue. SMQ does not maintain a hit count, instead it swaps hit entries with the least recently used entry from the level above. The over all ordering being a side effect of this stochastic process. With this scheme we can decide how many entries occupy each multiqueue level, resulting in better promotion/demotion decisions. Adaptability ------------ The MQ policy maintains a hit count for each cache block. For a different block to get promoted to the cache it's hit count has to exceed the lowest currently in the cache. This means it can take a long time for the cache to adapt between varying IO patterns. Periodically degrading the hit counts could help with this, but I haven't found a nice general solution. SMQ doesn't maintain hit counts, so a lot of this problem just goes away. In addition it tracks performance of the hotspot queue, which is used to decide which blocks to promote. If the hotspot queue is performing badly then it starts moving entries more quickly between levels. This lets it adapt to new IO patterns very quickly. Performance ----------- In my tests SMQ shows substantially better performance than MQ. Once this matures a bit more I'm sure it'll become the default policy. Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
2015-05-15 22:33:34 +08:00
}
module_init(smq_init);
module_exit(smq_exit);
MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
MODULE_LICENSE("GPL");
MODULE_DESCRIPTION("smq cache policy");
MODULE_ALIAS("dm-cache-default");
MODULE_ALIAS("dm-cache-mq");
MODULE_ALIAS("dm-cache-cleaner");