The precision loss of reading IO start_time with jiffies_to_nsecs
instead of using a high resolution timer degrades HST path prediction
for BIO-based mpath on high load workloads.
Below, I show the utilization percentage of a 10 disk multipath with
asymmetrical disk access cost, while being exercised by a randwrite FIO
benchmark with high submission queue depth (depth=64). It is possible
to see that the HST path selection degrades heavily for high-iops in
BIO-mpath, underutilizing the slower paths way beyond expected. This
seems to be caused by the start_time truncation, which makes some IO to
seem much slower than it actually is. In this scenario ST outperforms
HST for bio-mpath, but not for mq-mpath, which already uses ktime_get_ns().
The third column shows utilization with this patch applied. It is easy
to see that now HST prediction is much closer to the ideal distribution
(calculated considering the real cost of each path).
| | ST | HST (orig) | HST(ktime) | Best |
| sdd | 0.17 | 0.20 | 0.17 | 0.18 |
| sde | 0.17 | 0.20 | 0.17 | 0.18 |
| sdf | 0.17 | 0.20 | 0.17 | 0.18 |
| sdg | 0.06 | 0.00 | 0.06 | 0.04 |
| sdh | 0.03 | 0.00 | 0.03 | 0.02 |
| sdi | 0.03 | 0.00 | 0.03 | 0.02 |
| sdj | 0.02 | 0.00 | 0.01 | 0.01 |
| sdk | 0.02 | 0.00 | 0.01 | 0.01 |
| sdl | 0.17 | 0.20 | 0.17 | 0.18 |
| sdm | 0.17 | 0.20 | 0.17 | 0.18 |
This issue was originally discussed [1] when we first merged HST, and
this patch was left as a low hanging fruit to be solved later.
Regarding the implementation, as suggested by Mike in that mail thread,
in order to avoid the overhead of ktime_get_ns for other selectors, this
patch adds a flag for the selector code to request the high-resolution
timer.
I tested this using the same benchmark used in the original HST submission.
Full test and benchmark scripts are available here:
https://people.collabora.com/~krisman/HST-BIO-MPATH/
[1] https://lore.kernel.org/lkml/85tv0am9de.fsf@collabora.com/T/
Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.com>
[snitzer: cleaned up various implementation details]
Signed-off-by: Mike Snitzer <snitzer@kernel.org>
The HST path selector needs this information to perform path
prediction. For request-based mpath, struct request's io_start_time_ns
is used, while for bio-based, use the start_time stored in dm_io.
Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
If a path selector has any use for a repeat_count it should be handled
locally and not depend on the dm-mpath core to be concerned with it.
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
This patch makes two additions to the dm path selector interface for
dynamic load balancers:
o a new hook, start_io()
o a new parameter 'nr_bytes' to select_path()/start_io()/end_io()
to pass the size of the I/O
start_io() is called when a target driver actually submits I/O
to the selected path.
Path selectors can use it to start accounting of the I/O.
(e.g. counting the number of in-flight I/Os.)
The start_io hook is based on the patch posted by Stefan Bader:
https://www.redhat.com/archives/dm-devel/2005-October/msg00050.html
nr_bytes, the size of the I/O, is so path selectors can take the
size of the I/O into account when deciding which path to use.
dm-service-time uses it to estimate service time, for example.
(Added the nr_bytes member to dm_mpath_io instead of using existing
details.bi_size, since request-based dm patch deletes it.)
Signed-off-by: Stefan Bader <stefan.bader@canonical.com>
Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Rename DM's struct path to struct dm_path to prevent name collision between it
and struct path from fs/namei.c.
Signed-off-by: Josef "Jeff" Sipek <jsipek@cs.sunysb.edu>
Acked-by: Alasdair G Kergon <agk@redhat.com>
Cc: <reiserfs-dev@namesys.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.
Let it rip!
Gabriel Krisman Bertazi