[Problem Statement]
select_idle_cpu() might spend too much time searching for an idle CPU,
when the system is overloaded.
The following histogram is the time spent in select_idle_cpu(),
when running 224 instances of netperf on a system with 112 CPUs
per LLC domain:
@usecs:
[0] 533 | |
[1] 5495 | |
[2, 4) 12008 | |
[4, 8) 239252 | |
[8, 16) 4041924 |@@@@@@@@@@@@@@ |
[16, 32) 12357398 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ |
[32, 64) 14820255 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@|
[64, 128) 13047682 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ |
[128, 256) 8235013 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@ |
[256, 512) 4507667 |@@@@@@@@@@@@@@@ |
[512, 1K) 2600472 |@@@@@@@@@ |
[1K, 2K) 927912 |@@@ |
[2K, 4K) 218720 | |
[4K, 8K) 98161 | |
[8K, 16K) 37722 | |
[16K, 32K) 6715 | |
[32K, 64K) 477 | |
[64K, 128K) 7 | |
netperf latency usecs:
=======
case load Lat_99th std%
TCP_RR thread-224 257.39 ( 0.21)
The time spent in select_idle_cpu() is visible to netperf and might have a negative
impact.
[Symptom analysis]
The patch [1] from Mel Gorman has been applied to track the efficiency
of select_idle_sibling. Copy the indicators here:
SIS Search Efficiency(se_eff%):
A ratio expressed as a percentage of runqueues scanned versus
idle CPUs found. A 100% efficiency indicates that the target,
prev or recent CPU of a task was idle at wakeup. The lower the
efficiency, the more runqueues were scanned before an idle CPU
was found.
SIS Domain Search Efficiency(dom_eff%):
Similar, except only for the slower SIS
patch.
SIS Fast Success Rate(fast_rate%):
Percentage of SIS that used target, prev or
recent CPUs.
SIS Success rate(success_rate%):
Percentage of scans that found an idle CPU.
The test is based on Aubrey's schedtests tool, including netperf, hackbench,
schbench and tbench.
Test on vanilla kernel:
schedstat_parse.py -f netperf_vanilla.log
case load se_eff% dom_eff% fast_rate% success_rate%
TCP_RR 28 threads 99.978 18.535 99.995 100.000
TCP_RR 56 threads 99.397 5.671 99.964 100.000
TCP_RR 84 threads 21.721 6.818 73.632 100.000
TCP_RR 112 threads 12.500 5.533 59.000 100.000
TCP_RR 140 threads 8.524 4.535 49.020 100.000
TCP_RR 168 threads 6.438 3.945 40.309 99.999
TCP_RR 196 threads 5.397 3.718 32.320 99.982
TCP_RR 224 threads 4.874 3.661 25.775 99.767
UDP_RR 28 threads 99.988 17.704 99.997 100.000
UDP_RR 56 threads 99.528 5.977 99.970 100.000
UDP_RR 84 threads 24.219 6.992 76.479 100.000
UDP_RR 112 threads 13.907 5.706 62.538 100.000
UDP_RR 140 threads 9.408 4.699 52.519 100.000
UDP_RR 168 threads 7.095 4.077 44.352 100.000
UDP_RR 196 threads 5.757 3.775 35.764 99.991
UDP_RR 224 threads 5.124 3.704 28.748 99.860
schedstat_parse.py -f schbench_vanilla.log
(each group has 28 tasks)
case load se_eff% dom_eff% fast_rate% success_rate%
normal 1 mthread 99.152 6.400 99.941 100.000
normal 2 mthreads 97.844 4.003 99.908 100.000
normal 3 mthreads 96.395 2.118 99.917 99.998
normal 4 mthreads 55.288 1.451 98.615 99.804
normal 5 mthreads 7.004 1.870 45.597 61.036
normal 6 mthreads 3.354 1.346 20.777 34.230
normal 7 mthreads 2.183 1.028 11.257 21.055
normal 8 mthreads 1.653 0.825 7.849 15.549
schedstat_parse.py -f hackbench_vanilla.log
(each group has 28 tasks)
case load se_eff% dom_eff% fast_rate% success_rate%
process-pipe 1 group 99.991 7.692 99.999 100.000
process-pipe 2 groups 99.934 4.615 99.997 100.000
process-pipe 3 groups 99.597 3.198 99.987 100.000
process-pipe 4 groups 98.378 2.464 99.958 100.000
process-pipe 5 groups 27.474 3.653 89.811 99.800
process-pipe 6 groups 20.201 4.098 82.763 99.570
process-pipe 7 groups 16.423 4.156 77.398 99.316
process-pipe 8 groups 13.165 3.920 72.232 98.828
process-sockets 1 group 99.977 5.882 99.999 100.000
process-sockets 2 groups 99.927 5.505 99.996 100.000
process-sockets 3 groups 99.397 3.250 99.980 100.000
process-sockets 4 groups 79.680 4.258 98.864 99.998
process-sockets 5 groups 7.673 2.503 63.659 92.115
process-sockets 6 groups 4.642 1.584 58.946 88.048
process-sockets 7 groups 3.493 1.379 49.816 81.164
process-sockets 8 groups 3.015 1.407 40.845 75.500
threads-pipe 1 group 99.997 0.000 100.000 100.000
threads-pipe 2 groups 99.894 2.932 99.997 100.000
threads-pipe 3 groups 99.611 4.117 99.983 100.000
threads-pipe 4 groups 97.703 2.624 99.937 100.000
threads-pipe 5 groups 22.919 3.623 87.150 99.764
threads-pipe 6 groups 18.016 4.038 80.491 99.557
threads-pipe 7 groups 14.663 3.991 75.239 99.247
threads-pipe 8 groups 12.242 3.808 70.651 98.644
threads-sockets 1 group 99.990 6.667 99.999 100.000
threads-sockets 2 groups 99.940 5.114 99.997 100.000
threads-sockets 3 groups 99.469 4.115 99.977 100.000
threads-sockets 4 groups 87.528 4.038 99.400 100.000
threads-sockets 5 groups 6.942 2.398 59.244 88.337
threads-sockets 6 groups 4.359 1.954 49.448 87.860
threads-sockets 7 groups 2.845 1.345 41.198 77.102
threads-sockets 8 groups 2.871 1.404 38.512 74.312
schedstat_parse.py -f tbench_vanilla.log
case load se_eff% dom_eff% fast_rate% success_rate%
loopback 28 threads 99.976 18.369 99.995 100.000
loopback 56 threads 99.222 7.799 99.934 100.000
loopback 84 threads 19.723 6.819 70.215 100.000
loopback 112 threads 11.283 5.371 55.371 99.999
loopback 140 threads 0.000 0.000 0.000 0.000
loopback 168 threads 0.000 0.000 0.000 0.000
loopback 196 threads 0.000 0.000 0.000 0.000
loopback 224 threads 0.000 0.000 0.000 0.000
According to the test above, if the system becomes busy, the
SIS Search Efficiency(se_eff%) drops significantly. Although some
benchmarks would finally find an idle CPU(success_rate% = 100%), it is
doubtful whether it is worth it to search the whole LLC domain.
[Proposal]
It would be ideal to have a crystal ball to answer this question:
How many CPUs must a wakeup path walk down, before it can find an idle
CPU? Many potential metrics could be used to predict the number.
One candidate is the sum of util_avg in this LLC domain. The benefit
of choosing util_avg is that it is a metric of accumulated historic
activity, which seems to be smoother than instantaneous metrics
(such as rq->nr_running). Besides, choosing the sum of util_avg
would help predict the load of the LLC domain more precisely, because
SIS_PROP uses one CPU's idle time to estimate the total LLC domain idle
time.
In summary, the lower the util_avg is, the more select_idle_cpu()
should scan for idle CPU, and vice versa. When the sum of util_avg
in this LLC domain hits 85% or above, the scan stops. The reason to
choose 85% as the threshold is that this is the imbalance_pct(117)
when a LLC sched group is overloaded.
Introduce the quadratic function:
y = SCHED_CAPACITY_SCALE - p * x^2
and y'= y / SCHED_CAPACITY_SCALE
x is the ratio of sum_util compared to the CPU capacity:
x = sum_util / (llc_weight * SCHED_CAPACITY_SCALE)
y' is the ratio of CPUs to be scanned in the LLC domain,
and the number of CPUs to scan is calculated by:
nr_scan = llc_weight * y'
Choosing quadratic function is because:
[1] Compared to the linear function, it scans more aggressively when the
sum_util is low.
[2] Compared to the exponential function, it is easier to calculate.
[3] It seems that there is no accurate mapping between the sum of util_avg
and the number of CPUs to be scanned. Use heuristic scan for now.
For a platform with 112 CPUs per LLC, the number of CPUs to scan is:
sum_util% 0 5 15 25 35 45 55 65 75 85 86 ...
scan_nr 112 111 108 102 93 81 65 47 25 1 0 ...
For a platform with 16 CPUs per LLC, the number of CPUs to scan is:
sum_util% 0 5 15 25 35 45 55 65 75 85 86 ...
scan_nr 16 15 15 14 13 11 9 6 3 0 0 ...
Furthermore, to minimize the overhead of calculating the metrics in
select_idle_cpu(), borrow the statistics from periodic load balance.
As mentioned by Abel, on a platform with 112 CPUs per LLC, the
sum_util calculated by periodic load balance after 112 ms would
decay to about 0.5 * 0.5 * 0.5 * 0.7 = 8.75%, thus bringing a delay
in reflecting the latest utilization. But it is a trade-off.
Checking the util_avg in newidle load balance would be more frequent,
but it brings overhead - multiple CPUs write/read the per-LLC shared
variable and introduces cache contention. Tim also mentioned that,
it is allowed to be non-optimal in terms of scheduling for the
short-term variations, but if there is a long-term trend in the load
behavior, the scheduler can adjust for that.
When SIS_UTIL is enabled, the select_idle_cpu() uses the nr_scan
calculated by SIS_UTIL instead of the one from SIS_PROP. As Peter and
Mel suggested, SIS_UTIL should be enabled by default.
This patch is based on the util_avg, which is very sensitive to the
CPU frequency invariance. There is an issue that, when the max frequency
has been clamp, the util_avg would decay insanely fast when
the CPU is idle. Commit addca28512 ("cpufreq: intel_pstate: Handle no_turbo
in frequency invariance") could be used to mitigate this symptom, by adjusting
the arch_max_freq_ratio when turbo is disabled. But this issue is still
not thoroughly fixed, because the current code is unaware of the user-specified
max CPU frequency.
[Test result]
netperf and tbench were launched with 25% 50% 75% 100% 125% 150%
175% 200% of CPU number respectively. Hackbench and schbench were launched
by 1, 2 ,4, 8 groups. Each test lasts for 100 seconds and repeats 3 times.
The following is the benchmark result comparison between
baseline:vanilla v5.19-rc1 and compare:patched kernel. Positive compare%
indicates better performance.
Each netperf test is a:
netperf -4 -H 127.0.1 -t TCP/UDP_RR -c -C -l 100
netperf.throughput
=======
case load baseline(std%) compare%( std%)
TCP_RR 28 threads 1.00 ( 0.34) -0.16 ( 0.40)
TCP_RR 56 threads 1.00 ( 0.19) -0.02 ( 0.20)
TCP_RR 84 threads 1.00 ( 0.39) -0.47 ( 0.40)
TCP_RR 112 threads 1.00 ( 0.21) -0.66 ( 0.22)
TCP_RR 140 threads 1.00 ( 0.19) -0.69 ( 0.19)
TCP_RR 168 threads 1.00 ( 0.18) -0.48 ( 0.18)
TCP_RR 196 threads 1.00 ( 0.16) +194.70 ( 16.43)
TCP_RR 224 threads 1.00 ( 0.16) +197.30 ( 7.85)
UDP_RR 28 threads 1.00 ( 0.37) +0.35 ( 0.33)
UDP_RR 56 threads 1.00 ( 11.18) -0.32 ( 0.21)
UDP_RR 84 threads 1.00 ( 1.46) -0.98 ( 0.32)
UDP_RR 112 threads 1.00 ( 28.85) -2.48 ( 19.61)
UDP_RR 140 threads 1.00 ( 0.70) -0.71 ( 14.04)
UDP_RR 168 threads 1.00 ( 14.33) -0.26 ( 11.16)
UDP_RR 196 threads 1.00 ( 12.92) +186.92 ( 20.93)
UDP_RR 224 threads 1.00 ( 11.74) +196.79 ( 18.62)
Take the 224 threads as an example, the SIS search metrics changes are
illustrated below:
vanilla patched
4544492 +237.5% 15338634 sched_debug.cpu.sis_domain_search.avg
38539 +39686.8% 15333634 sched_debug.cpu.sis_failed.avg
128300000 -87.9% 15551326 sched_debug.cpu.sis_scanned.avg
5842896 +162.7% 15347978 sched_debug.cpu.sis_search.avg
There is -87.9% less CPU scans after patched, which indicates lower overhead.
Besides, with this patch applied, there is -13% less rq lock contention
in perf-profile.calltrace.cycles-pp._raw_spin_lock.raw_spin_rq_lock_nested
.try_to_wake_up.default_wake_function.woken_wake_function.
This might help explain the performance improvement - Because this patch allows
the waking task to remain on the previous CPU, rather than grabbing other CPUs'
lock.
Each hackbench test is a:
hackbench -g $job --process/threads --pipe/sockets -l 1000000 -s 100
hackbench.throughput
=========
case load baseline(std%) compare%( std%)
process-pipe 1 group 1.00 ( 1.29) +0.57 ( 0.47)
process-pipe 2 groups 1.00 ( 0.27) +0.77 ( 0.81)
process-pipe 4 groups 1.00 ( 0.26) +1.17 ( 0.02)
process-pipe 8 groups 1.00 ( 0.15) -4.79 ( 0.02)
process-sockets 1 group 1.00 ( 0.63) -0.92 ( 0.13)
process-sockets 2 groups 1.00 ( 0.03) -0.83 ( 0.14)
process-sockets 4 groups 1.00 ( 0.40) +5.20 ( 0.26)
process-sockets 8 groups 1.00 ( 0.04) +3.52 ( 0.03)
threads-pipe 1 group 1.00 ( 1.28) +0.07 ( 0.14)
threads-pipe 2 groups 1.00 ( 0.22) -0.49 ( 0.74)
threads-pipe 4 groups 1.00 ( 0.05) +1.88 ( 0.13)
threads-pipe 8 groups 1.00 ( 0.09) -4.90 ( 0.06)
threads-sockets 1 group 1.00 ( 0.25) -0.70 ( 0.53)
threads-sockets 2 groups 1.00 ( 0.10) -0.63 ( 0.26)
threads-sockets 4 groups 1.00 ( 0.19) +11.92 ( 0.24)
threads-sockets 8 groups 1.00 ( 0.08) +4.31 ( 0.11)
Each tbench test is a:
tbench -t 100 $job 127.0.0.1
tbench.throughput
======
case load baseline(std%) compare%( std%)
loopback 28 threads 1.00 ( 0.06) -0.14 ( 0.09)
loopback 56 threads 1.00 ( 0.03) -0.04 ( 0.17)
loopback 84 threads 1.00 ( 0.05) +0.36 ( 0.13)
loopback 112 threads 1.00 ( 0.03) +0.51 ( 0.03)
loopback 140 threads 1.00 ( 0.02) -1.67 ( 0.19)
loopback 168 threads 1.00 ( 0.38) +1.27 ( 0.27)
loopback 196 threads 1.00 ( 0.11) +1.34 ( 0.17)
loopback 224 threads 1.00 ( 0.11) +1.67 ( 0.22)
Each schbench test is a:
schbench -m $job -t 28 -r 100 -s 30000 -c 30000
schbench.latency_90%_us
========
case load baseline(std%) compare%( std%)
normal 1 mthread 1.00 ( 31.22) -7.36 ( 20.25)*
normal 2 mthreads 1.00 ( 2.45) -0.48 ( 1.79)
normal 4 mthreads 1.00 ( 1.69) +0.45 ( 0.64)
normal 8 mthreads 1.00 ( 5.47) +9.81 ( 14.28)
*Consider the Standard Deviation, this -7.36% regression might not be valid.
Also, a OLTP workload with a commercial RDBMS has been tested, and there
is no significant change.
There were concerns that unbalanced tasks among CPUs would cause problems.
For example, suppose the LLC domain is composed of 8 CPUs, and 7 tasks are
bound to CPU0~CPU6, while CPU7 is idle:
CPU0 CPU1 CPU2 CPU3 CPU4 CPU5 CPU6 CPU7
util_avg 1024 1024 1024 1024 1024 1024 1024 0
Since the util_avg ratio is 87.5%( = 7/8 ), which is higher than 85%,
select_idle_cpu() will not scan, thus CPU7 is undetected during scan.
But according to Mel, it is unlikely the CPU7 will be idle all the time
because CPU7 could pull some tasks via CPU_NEWLY_IDLE.
lkp(kernel test robot) has reported a regression on stress-ng.sock on a
very busy system. According to the sched_debug statistics, it might be caused
by SIS_UTIL terminates the scan and chooses a previous CPU earlier, and this
might introduce more context switch, especially involuntary preemption, which
impacts a busy stress-ng. This regression has shown that, not all benchmarks
in every scenario benefit from idle CPU scan limit, and it needs further
investigation.
Besides, there is slight regression in hackbench's 16 groups case when the
LLC domain has 16 CPUs. Prateek mentioned that we should scan aggressively
in an LLC domain with 16 CPUs. Because the cost to search for an idle one
among 16 CPUs is negligible. The current patch aims to propose a generic
solution and only considers the util_avg. Something like the below could
be applied on top of the current patch to fulfill the requirement:
if (llc_weight <= 16)
nr_scan = nr_scan * 32 / llc_weight;
For LLC domain with 16 CPUs, the nr_scan will be expanded to 2 times large.
The smaller the CPU number this LLC domain has, the larger nr_scan will be
expanded. This needs further investigation.
There is also ongoing work[2] from Abel to filter out the busy CPUs during
wakeup, to further speed up the idle CPU scan. And it could be a following-up
optimization on top of this change.
Suggested-by: Tim Chen <tim.c.chen@intel.com>
Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Chen Yu <yu.c.chen@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Yicong Yang <yangyicong@hisilicon.com>
Tested-by: Mohini Narkhede <mohini.narkhede@intel.com>
Tested-by: K Prateek Nayak <kprateek.nayak@amd.com>
Link: https://lore.kernel.org/r/20220612163428.849378-1-yu.c.chen@intel.com
The queued remote wakeup mechanism has turned out to be suboptimal for RT
enabled kernels. The maximum latencies go up by a factor of > 5x in certain
scenarious.
This is caused by either long wake lists or by a large number of TTWU IPIs
which are processed back to back.
Disable it for RT.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20210928122411.482262764@linutronix.de
CPU scheduler marks need_resched flag to signal a schedule() on a
particular CPU. But, schedule() may not happen immediately in cases
where the current task is executing in the kernel mode (no
preemption state) for extended periods of time.
This patch adds a warn_on if need_resched is pending for more than the
time specified in sysctl resched_latency_warn_ms. If it goes off, it is
likely that there is a missing cond_resched() somewhere. Monitoring is
done via the tick and the accuracy is hence limited to jiffy scale. This
also means that we won't trigger the warning if the tick is disabled.
This feature (LATENCY_WARN) is default disabled.
Signed-off-by: Paul Turner <pjt@google.com>
Signed-off-by: Josh Don <joshdon@google.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20210416212936.390566-1-joshdon@google.com
The current sched_slice() seems to have issues; there's two possible
things that could be improved:
- the 'nr_running' used for __sched_period() is daft when cgroups are
considered. Using the RQ wide h_nr_running seems like a much more
consistent number.
- (esp) cgroups can slice it real fine, which makes for easy
over-scheduling, ensure min_gran is what the name says.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Valentin Schneider <valentin.schneider@arm.com>
Link: https://lkml.kernel.org/r/20210412102001.611897312@infradead.org
Fix ~42 single-word typos in scheduler code comments.
We have accumulated a few fun ones over the years. :-)
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Juri Lelli <juri.lelli@redhat.com>
Cc: Vincent Guittot <vincent.guittot@linaro.org>
Cc: Dietmar Eggemann <dietmar.eggemann@arm.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Ben Segall <bsegall@google.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: linux-kernel@vger.kernel.org
The HRTICK feature has traditionally been servicing configurations that
need precise preemptions point for NORMAL tasks. More recently, the
feature has been extended to also service DEADLINE tasks with stringent
runtime enforcement needs (e.g., runtime < 1ms with HZ=1000).
Enabling HRTICK sched feature currently enables the additional timer and
task tick for both classes, which might introduced undesired overhead
for no additional benefit if one needed it only for one of the cases.
Separate HRTICK sched feature in two (and leave the traditional case
name unmodified) so that it can be selectively enabled when needed.
With:
$ echo HRTICK > /sys/kernel/debug/sched_features
the NORMAL/fair hrtick gets enabled.
With:
$ echo HRTICK_DL > /sys/kernel/debug/sched_features
the DEADLINE hrtick gets enabled.
Signed-off-by: Juri Lelli <juri.lelli@redhat.com>
Signed-off-by: Luis Claudio R. Goncalves <lgoncalv@redhat.com>
Signed-off-by: Daniel Bristot de Oliveira <bristot@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lkml.kernel.org/r/20210208073554.14629-3-juri.lelli@redhat.com
SIS_AVG_CPU was introduced as a means of avoiding a search when the
average search cost indicated that the search would likely fail. It was
a blunt instrument and disabled by commit 4c77b18cf8 ("sched/fair: Make
select_idle_cpu() more aggressive") and later replaced with a proportional
search depth by commit 1ad3aaf3fc ("sched/core: Implement new approach
to scale select_idle_cpu()").
While there are corner cases where SIS_AVG_CPU is better, it has now been
disabled for almost three years. As the intent of SIS_PROP is to reduce
the time complexity of select_idle_cpu(), lets drop SIS_AVG_CPU and focus
on SIS_PROP as a throttling mechanism.
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org>
Link: https://lkml.kernel.org/r/20210125085909.4600-2-mgorman@techsingularity.net
The RT_RUNTIME_SHARE sched feature enables the sharing of rt_runtime
between CPUs, allowing a CPU to run a real-time task up to 100% of the
time while leaving more space for non-real-time tasks to run on the CPU
that lend rt_runtime.
The problem is that a CPU can easily borrow enough rt_runtime to allow
a spinning rt-task to run forever, starving per-cpu tasks like kworkers,
which are non-real-time by design.
This patch disables RT_RUNTIME_SHARE by default, avoiding this problem.
The feature will still be present for users that want to enable it,
though.
Signed-off-by: Daniel Bristot de Oliveira <bristot@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Wei Wang <wvw@google.com>
Link: https://lkml.kernel.org/r/b776ab46817e3db5d8ef79175fa0d71073c051c7.1600697903.git.bristot@redhat.com
The estimated utilization for a task:
util_est = max(util_avg, est.enqueue, est.ewma)
is defined based on:
- util_avg: the PELT defined utilization
- est.enqueued: the util_avg at the end of the last activation
- est.ewma: a exponential moving average on the est.enqueued samples
According to this definition, when a task suddenly changes its bandwidth
requirements from small to big, the EWMA will need to collect multiple
samples before converging up to track the new big utilization.
This slow convergence towards bigger utilization values is not
aligned to the default scheduler behavior, which is to optimize for
performance. Moreover, the est.ewma component fails to compensate for
temporarely utilization drops which spans just few est.enqueued samples.
To let util_est do a better job in the scenario depicted above, change
its definition by making util_est directly follow upward motion and
only decay the est.ewma on downward.
Signed-off-by: Patrick Bellasi <patrick.bellasi@matbug.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Vincent Guittot <vincent.guittot@linaro.org>
Cc: Dietmar Eggemann <dietmar.eggemann@arm.com>
Cc: Douglas Raillard <douglas.raillard@arm.com>
Cc: Juri Lelli <juri.lelli@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Quentin Perret <qperret@google.com>
Cc: Rafael J . Wysocki <rafael.j.wysocki@intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: https://lkml.kernel.org/r/20191023205630.14469-1-patrick.bellasi@matbug.net
Signed-off-by: Ingo Molnar <mingo@kernel.org>
With LB_BIAS disabled, source_load() & target_load() return
weighted_cpuload(). Replace both with calls to weighted_cpuload().
The function to obtain the load index (sd->*_idx) for an sd,
get_sd_load_idx(), can be removed as well.
Finally, get rid of the sched feature LB_BIAS.
Signed-off-by: Dietmar Eggemann <dietmar.eggemann@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Rik van Riel <riel@surriel.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Morten Rasmussen <morten.rasmussen@arm.com>
Cc: Patrick Bellasi <patrick.bellasi@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Quentin Perret <quentin.perret@arm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Valentin Schneider <valentin.schneider@arm.com>
Cc: Vincent Guittot <vincent.guittot@linaro.org>
Link: https://lkml.kernel.org/r/20190527062116.11512-3-dietmar.eggemann@arm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
LB_BIAS allows the adjustment on how conservative load should be
balanced.
The rq->cpu_load[idx] array is used for this functionality. It contains
weighted CPU load decayed average values over different intervals
(idx = 1..4). Idx = 0 is the weighted CPU load itself.
The values are updated during scheduler_tick, before idle balance and at
nohz exit.
There are 5 different types of idx's per sched domain (sd). Each of them
is used to index into the rq->cpu_load[idx] array in a specific scenario
(busy, idle and newidle for load balancing, forkexec for wake-up
slow-path load balancing and wake for affine wakeup based on weight).
Only the sd idx's for busy and idle load balancing are set to 2,3 or 1,2
respectively. All the other sd idx's are set to 0.
Conservative load balancing is achieved for sd idx's >= 1 by using the
min/max (source_load()/target_load()) value between the current weighted
CPU load and the rq->cpu_load[sd idx -1] for the busiest(idlest)/local
CPU load in load balancing or vice versa in the wake-up slow-path load
balancing.
There is no conservative balancing for sd idx = 0 since only current
weighted CPU load is used in this case.
It is very likely that LB_BIAS' influence on load balancing can be
neglected (see test results below). This is further supported by:
(1) Weighted CPU load today is by itself a decayed average value (PELT)
(cfs_rq->avg->runnable_load_avg) and not the instantaneous load
(rq->load.weight) it was when LB_BIAS was introduced.
(2) Sd imbalance_pct is used for CPU_NEWLY_IDLE and CPU_NOT_IDLE (relate
to sd's newidle and busy idx) in find_busiest_group() when comparing
busiest and local avg load to make load balancing even more
conservative.
(3) The sd forkexec and newidle idx are always set to 0 so there is no
adjustment on how conservatively load balancing is done here.
(4) Affine wakeup based on weight (wake_affine_weight()) will not be
impacted since the sd wake idx is always set to 0.
Let's disable LB_BIAS by default for a few kernel releases to make sure
that no workload and no scheduler topology is affected. The benefit of
being able to remove the LB_BIAS dependency from source_load() and
target_load() is that the entire rq->cpu_load[idx] code could be removed
in this case.
It is really hard to say if there is no regression w/o testing this with
a lot of different workloads on a lot of different platforms, especially
NUMA machines.
The following 104 LKP (Linux Kernel Performance) tests were run by the
0-Day guys mostly on multi-socket hosts with a larger number of logical
cpus (88, 192).
The base for the test was commit b3dae109fa ("sched/swait: Rename to
exclusive") (tip/sched/core v4.18-rc1).
Only 2 out of the 104 tests had a significant change in one of the
metrics (fsmark/1x-1t-1HDD-btrfs-nfsv4-4M-60G-NoSync-performance +7%
files_per_sec, unixbench/300s-100%-syscall-performance -11% score).
Tests which showed a change in one of the metrics are marked with a '*'
and this change is listed as well.
(a) lkp-bdw-ep3:
88 threads Intel(R) Xeon(R) CPU E5-2699 v4 @ 2.20GHz 64G
dd-write/10m-1HDD-cfq-btrfs-100dd-performance
fsmark/1x-1t-1HDD-xfs-nfsv4-4M-60G-NoSync-performance
* fsmark/1x-1t-1HDD-btrfs-nfsv4-4M-60G-NoSync-performance
7.50 7% 8.00 ± 6% fsmark.files_per_sec
fsmark/1x-1t-1HDD-btrfs-nfsv4-4M-60G-fsyncBeforeClose-performance
fsmark/1x-1t-1HDD-btrfs-4M-60G-NoSync-performance
fsmark/1x-1t-1HDD-btrfs-4M-60G-fsyncBeforeClose-performance
kbuild/300s-50%-vmlinux_prereq-performance
kbuild/300s-200%-vmlinux_prereq-performance
kbuild/300s-50%-vmlinux_prereq-performance-1HDD-ext4
kbuild/300s-200%-vmlinux_prereq-performance-1HDD-ext4
(b) lkp-skl-4sp1:
192 threads Intel(R) Xeon(R) Platinum 8160 768G
dbench/100%-performance
ebizzy/200%-100x-10s-performance
hackbench/1600%-process-pipe-performance
iperf/300s-cs-localhost-tcp-performance
iperf/300s-cs-localhost-udp-performance
perf-bench-numa-mem/2t-300M-performance
perf-bench-sched-pipe/10000000ops-process-performance
perf-bench-sched-pipe/10000000ops-threads-performance
schbench/2-16-300-30000-30000-performance
tbench/100%-cs-localhost-performance
(c) lkp-bdw-ep6:
88 threads Intel(R) Xeon(R) CPU E5-2699 v4 @ 2.20GHz 128G
stress-ng/100%-60s-pipe-performance
unixbench/300s-1-whetstone-double-performance
unixbench/300s-1-shell1-performance
unixbench/300s-1-shell8-performance
unixbench/300s-1-pipe-performance
* unixbench/300s-1-context1-performance
312 315 unixbench.score
unixbench/300s-1-spawn-performance
unixbench/300s-1-syscall-performance
unixbench/300s-1-dhry2reg-performance
unixbench/300s-1-fstime-performance
unixbench/300s-1-fsbuffer-performance
unixbench/300s-1-fsdisk-performance
unixbench/300s-100%-whetstone-double-performance
unixbench/300s-100%-shell1-performance
unixbench/300s-100%-shell8-performance
unixbench/300s-100%-pipe-performance
unixbench/300s-100%-context1-performance
unixbench/300s-100%-spawn-performance
* unixbench/300s-100%-syscall-performance
3571 ± 3% -11% 3183 ± 4% unixbench.score
unixbench/300s-100%-dhry2reg-performance
unixbench/300s-100%-fstime-performance
unixbench/300s-100%-fsbuffer-performance
unixbench/300s-100%-fsdisk-performance
unixbench/300s-1-execl-performance
unixbench/300s-100%-execl-performance
* will-it-scale/brk1-performance
365004 360387 will-it-scale.per_thread_ops
* will-it-scale/dup1-performance
432401 437596 will-it-scale.per_thread_ops
will-it-scale/eventfd1-performance
will-it-scale/futex1-performance
will-it-scale/futex2-performance
will-it-scale/futex3-performance
will-it-scale/futex4-performance
will-it-scale/getppid1-performance
will-it-scale/lock1-performance
will-it-scale/lseek1-performance
will-it-scale/lseek2-performance
* will-it-scale/malloc1-performance
47025 45817 will-it-scale.per_thread_ops
77499 76529 will-it-scale.per_process_ops
will-it-scale/malloc2-performance
* will-it-scale/mmap1-performance
123399 120815 will-it-scale.per_thread_ops
152219 149833 will-it-scale.per_process_ops
* will-it-scale/mmap2-performance
107327 104714 will-it-scale.per_thread_ops
136405 133765 will-it-scale.per_process_ops
will-it-scale/open1-performance
* will-it-scale/open2-performance
171570 168805 will-it-scale.per_thread_ops
532644 526202 will-it-scale.per_process_ops
will-it-scale/page_fault1-performance
will-it-scale/page_fault2-performance
will-it-scale/page_fault3-performance
will-it-scale/pipe1-performance
will-it-scale/poll1-performance
* will-it-scale/poll2-performance
176134 172848 will-it-scale.per_thread_ops
281361 275053 will-it-scale.per_process_ops
will-it-scale/posix_semaphore1-performance
will-it-scale/pread1-performance
will-it-scale/pread2-performance
will-it-scale/pread3-performance
will-it-scale/pthread_mutex1-performance
will-it-scale/pthread_mutex2-performance
will-it-scale/pwrite1-performance
will-it-scale/pwrite2-performance
will-it-scale/pwrite3-performance
* will-it-scale/read1-performance
1190563 1174833 will-it-scale.per_thread_ops
* will-it-scale/read2-performance
1105369 1080427 will-it-scale.per_thread_ops
will-it-scale/readseek1-performance
* will-it-scale/readseek2-performance
261818 259040 will-it-scale.per_thread_ops
will-it-scale/readseek3-performance
* will-it-scale/sched_yield-performance
2408059 2382034 will-it-scale.per_thread_ops
will-it-scale/signal1-performance
will-it-scale/unix1-performance
will-it-scale/unlink1-performance
will-it-scale/unlink2-performance
* will-it-scale/write1-performance
976701 961588 will-it-scale.per_thread_ops
* will-it-scale/writeseek1-performance
831898 822448 will-it-scale.per_thread_ops
* will-it-scale/writeseek2-performance
228248 225065 will-it-scale.per_thread_ops
* will-it-scale/writeseek3-performance
226670 224058 will-it-scale.per_thread_ops
will-it-scale/context_switch1-performance
aim7/performance-fork_test-2000
* aim7/performance-brk_test-3000
74869 76676 aim7.jobs-per-min
aim7/performance-disk_cp-3000
aim7/performance-disk_rd-3000
aim7/performance-sieve-3000
aim7/performance-page_test-3000
aim7/performance-creat-clo-3000
aim7/performance-mem_rtns_1-8000
aim7/performance-disk_wrt-8000
aim7/performance-pipe_cpy-8000
aim7/performance-ram_copy-8000
(d) lkp-avoton3:
8 threads Intel(R) Atom(TM) CPU C2750 @ 2.40GHz 16G
netperf/ipv4-900s-200%-cs-localhost-TCP_STREAM-performance
Signed-off-by: Dietmar Eggemann <dietmar.eggemann@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Fengguang Wu <fengguang.wu@intel.com>
Cc: Li Zhijian <zhijianx.li@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20180809135753.21077-1-dietmar.eggemann@arm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The estimated utilization of a task is currently updated every time the
task is dequeued. However, to keep overheads under control, PELT signals
are effectively updated at maximum once every 1ms.
Thus, for really short running tasks, it can happen that their util_avg
value has not been updates since their last enqueue. If such tasks are
also frequently running tasks (e.g. the kind of workload generated by
hackbench) it can also happen that their util_avg is updated only every
few activations.
This means that updating util_est at every dequeue potentially introduces
not necessary overheads and it's also conceptually wrong if the util_avg
signal has never been updated during a task activation.
Let's introduce a throttling mechanism on task's util_est updates
to sync them with util_avg updates. To make the solution memory
efficient, both in terms of space and load/store operations, we encode a
synchronization flag into the LSB of util_est.enqueued.
This makes util_est an even values only metric, which is still
considered good enough for its purpose.
The synchronization bit is (re)set by __update_load_avg_se() once the
PELT signal of a task has been updated during its last activation.
Such a throttling mechanism allows to keep under control util_est
overheads in the wakeup hot path, thus making it a suitable mechanism
which can be enabled also on high-intensity workload systems.
Thus, this now switches on by default the estimation utilization
scheduler feature.
Suggested-by: Chris Redpath <chris.redpath@arm.com>
Signed-off-by: Patrick Bellasi <patrick.bellasi@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Dietmar Eggemann <dietmar.eggemann@arm.com>
Cc: Joel Fernandes <joelaf@google.com>
Cc: Juri Lelli <juri.lelli@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Morten Rasmussen <morten.rasmussen@arm.com>
Cc: Paul Turner <pjt@google.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rafael J . Wysocki <rafael.j.wysocki@intel.com>
Cc: Steve Muckle <smuckle@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Todd Kjos <tkjos@android.com>
Cc: Vincent Guittot <vincent.guittot@linaro.org>
Cc: Viresh Kumar <viresh.kumar@linaro.org>
Link: http://lkml.kernel.org/r/20180309095245.11071-5-patrick.bellasi@arm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The util_avg signal computed by PELT is too variable for some use-cases.
For example, a big task waking up after a long sleep period will have its
utilization almost completely decayed. This introduces some latency before
schedutil will be able to pick the best frequency to run a task.
The same issue can affect task placement. Indeed, since the task
utilization is already decayed at wakeup, when the task is enqueued in a
CPU, this can result in a CPU running a big task as being temporarily
represented as being almost empty. This leads to a race condition where
other tasks can be potentially allocated on a CPU which just started to run
a big task which slept for a relatively long period.
Moreover, the PELT utilization of a task can be updated every [ms], thus
making it a continuously changing value for certain longer running
tasks. This means that the instantaneous PELT utilization of a RUNNING
task is not really meaningful to properly support scheduler decisions.
For all these reasons, a more stable signal can do a better job of
representing the expected/estimated utilization of a task/cfs_rq.
Such a signal can be easily created on top of PELT by still using it as
an estimator which produces values to be aggregated on meaningful
events.
This patch adds a simple implementation of util_est, a new signal built on
top of PELT's util_avg where:
util_est(task) = max(task::util_avg, f(task::util_avg@dequeue))
This allows to remember how big a task has been reported by PELT in its
previous activations via f(task::util_avg@dequeue), which is the new
_task_util_est(struct task_struct*) function added by this patch.
If a task should change its behavior and it runs longer in a new
activation, after a certain time its util_est will just track the
original PELT signal (i.e. task::util_avg).
The estimated utilization of cfs_rq is defined only for root ones.
That's because the only sensible consumer of this signal are the
scheduler and schedutil when looking for the overall CPU utilization
due to FAIR tasks.
For this reason, the estimated utilization of a root cfs_rq is simply
defined as:
util_est(cfs_rq) = max(cfs_rq::util_avg, cfs_rq::util_est::enqueued)
where:
cfs_rq::util_est::enqueued = sum(_task_util_est(task))
for each RUNNABLE task on that root cfs_rq
It's worth noting that the estimated utilization is tracked only for
objects of interests, specifically:
- Tasks: to better support tasks placement decisions
- root cfs_rqs: to better support both tasks placement decisions as
well as frequencies selection
Signed-off-by: Patrick Bellasi <patrick.bellasi@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com>
Cc: Joel Fernandes <joelaf@google.com>
Cc: Juri Lelli <juri.lelli@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Morten Rasmussen <morten.rasmussen@arm.com>
Cc: Paul Turner <pjt@google.com>
Cc: Rafael J . Wysocki <rafael.j.wysocki@intel.com>
Cc: Steve Muckle <smuckle@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Todd Kjos <tkjos@android.com>
Cc: Vincent Guittot <vincent.guittot@linaro.org>
Cc: Viresh Kumar <viresh.kumar@linaro.org>
Link: http://lkml.kernel.org/r/20180309095245.11071-2-patrick.bellasi@arm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Many source files in the tree are missing licensing information, which
makes it harder for compliance tools to determine the correct license.
By default all files without license information are under the default
license of the kernel, which is GPL version 2.
Update the files which contain no license information with the 'GPL-2.0'
SPDX license identifier. The SPDX identifier is a legally binding
shorthand, which can be used instead of the full boiler plate text.
This patch is based on work done by Thomas Gleixner and Kate Stewart and
Philippe Ombredanne.
How this work was done:
Patches were generated and checked against linux-4.14-rc6 for a subset of
the use cases:
- file had no licensing information it it.
- file was a */uapi/* one with no licensing information in it,
- file was a */uapi/* one with existing licensing information,
Further patches will be generated in subsequent months to fix up cases
where non-standard license headers were used, and references to license
had to be inferred by heuristics based on keywords.
The analysis to determine which SPDX License Identifier to be applied to
a file was done in a spreadsheet of side by side results from of the
output of two independent scanners (ScanCode & Windriver) producing SPDX
tag:value files created by Philippe Ombredanne. Philippe prepared the
base worksheet, and did an initial spot review of a few 1000 files.
The 4.13 kernel was the starting point of the analysis with 60,537 files
assessed. Kate Stewart did a file by file comparison of the scanner
results in the spreadsheet to determine which SPDX license identifier(s)
to be applied to the file. She confirmed any determination that was not
immediately clear with lawyers working with the Linux Foundation.
Criteria used to select files for SPDX license identifier tagging was:
- Files considered eligible had to be source code files.
- Make and config files were included as candidates if they contained >5
lines of source
- File already had some variant of a license header in it (even if <5
lines).
All documentation files were explicitly excluded.
The following heuristics were used to determine which SPDX license
identifiers to apply.
- when both scanners couldn't find any license traces, file was
considered to have no license information in it, and the top level
COPYING file license applied.
For non */uapi/* files that summary was:
SPDX license identifier # files
---------------------------------------------------|-------
GPL-2.0 11139
and resulted in the first patch in this series.
If that file was a */uapi/* path one, it was "GPL-2.0 WITH
Linux-syscall-note" otherwise it was "GPL-2.0". Results of that was:
SPDX license identifier # files
---------------------------------------------------|-------
GPL-2.0 WITH Linux-syscall-note 930
and resulted in the second patch in this series.
- if a file had some form of licensing information in it, and was one
of the */uapi/* ones, it was denoted with the Linux-syscall-note if
any GPL family license was found in the file or had no licensing in
it (per prior point). Results summary:
SPDX license identifier # files
---------------------------------------------------|------
GPL-2.0 WITH Linux-syscall-note 270
GPL-2.0+ WITH Linux-syscall-note 169
((GPL-2.0 WITH Linux-syscall-note) OR BSD-2-Clause) 21
((GPL-2.0 WITH Linux-syscall-note) OR BSD-3-Clause) 17
LGPL-2.1+ WITH Linux-syscall-note 15
GPL-1.0+ WITH Linux-syscall-note 14
((GPL-2.0+ WITH Linux-syscall-note) OR BSD-3-Clause) 5
LGPL-2.0+ WITH Linux-syscall-note 4
LGPL-2.1 WITH Linux-syscall-note 3
((GPL-2.0 WITH Linux-syscall-note) OR MIT) 3
((GPL-2.0 WITH Linux-syscall-note) AND MIT) 1
and that resulted in the third patch in this series.
- when the two scanners agreed on the detected license(s), that became
the concluded license(s).
- when there was disagreement between the two scanners (one detected a
license but the other didn't, or they both detected different
licenses) a manual inspection of the file occurred.
- In most cases a manual inspection of the information in the file
resulted in a clear resolution of the license that should apply (and
which scanner probably needed to revisit its heuristics).
- When it was not immediately clear, the license identifier was
confirmed with lawyers working with the Linux Foundation.
- If there was any question as to the appropriate license identifier,
the file was flagged for further research and to be revisited later
in time.
In total, over 70 hours of logged manual review was done on the
spreadsheet to determine the SPDX license identifiers to apply to the
source files by Kate, Philippe, Thomas and, in some cases, confirmation
by lawyers working with the Linux Foundation.
Kate also obtained a third independent scan of the 4.13 code base from
FOSSology, and compared selected files where the other two scanners
disagreed against that SPDX file, to see if there was new insights. The
Windriver scanner is based on an older version of FOSSology in part, so
they are related.
Thomas did random spot checks in about 500 files from the spreadsheets
for the uapi headers and agreed with SPDX license identifier in the
files he inspected. For the non-uapi files Thomas did random spot checks
in about 15000 files.
In initial set of patches against 4.14-rc6, 3 files were found to have
copy/paste license identifier errors, and have been fixed to reflect the
correct identifier.
Additionally Philippe spent 10 hours this week doing a detailed manual
inspection and review of the 12,461 patched files from the initial patch
version early this week with:
- a full scancode scan run, collecting the matched texts, detected
license ids and scores
- reviewing anything where there was a license detected (about 500+
files) to ensure that the applied SPDX license was correct
- reviewing anything where there was no detection but the patch license
was not GPL-2.0 WITH Linux-syscall-note to ensure that the applied
SPDX license was correct
This produced a worksheet with 20 files needing minor correction. This
worksheet was then exported into 3 different .csv files for the
different types of files to be modified.
These .csv files were then reviewed by Greg. Thomas wrote a script to
parse the csv files and add the proper SPDX tag to the file, in the
format that the file expected. This script was further refined by Greg
based on the output to detect more types of files automatically and to
distinguish between header and source .c files (which need different
comment types.) Finally Greg ran the script using the .csv files to
generate the patches.
Reviewed-by: Kate Stewart <kstewart@linuxfoundation.org>
Reviewed-by: Philippe Ombredanne <pombredanne@nexb.com>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The trivial wake_affine_idle() implementation is very good for a
number of workloads, but it comes apart at the moment there are no
idle CPUs left, IOW. the overloaded case.
hackbench:
NO_WA_WEIGHT WA_WEIGHT
hackbench-20 : 7.362717561 seconds 6.450509391 seconds
(win)
netperf:
NO_WA_WEIGHT WA_WEIGHT
TCP_SENDFILE-1 : Avg: 54524.6 Avg: 52224.3
TCP_SENDFILE-10 : Avg: 48185.2 Avg: 46504.3
TCP_SENDFILE-20 : Avg: 29031.2 Avg: 28610.3
TCP_SENDFILE-40 : Avg: 9819.72 Avg: 9253.12
TCP_SENDFILE-80 : Avg: 5355.3 Avg: 4687.4
TCP_STREAM-1 : Avg: 41448.3 Avg: 42254
TCP_STREAM-10 : Avg: 24123.2 Avg: 25847.9
TCP_STREAM-20 : Avg: 15834.5 Avg: 18374.4
TCP_STREAM-40 : Avg: 5583.91 Avg: 5599.57
TCP_STREAM-80 : Avg: 2329.66 Avg: 2726.41
TCP_RR-1 : Avg: 80473.5 Avg: 82638.8
TCP_RR-10 : Avg: 72660.5 Avg: 73265.1
TCP_RR-20 : Avg: 52607.1 Avg: 52634.5
TCP_RR-40 : Avg: 57199.2 Avg: 56302.3
TCP_RR-80 : Avg: 25330.3 Avg: 26867.9
UDP_RR-1 : Avg: 108266 Avg: 107844
UDP_RR-10 : Avg: 95480 Avg: 95245.2
UDP_RR-20 : Avg: 68770.8 Avg: 68673.7
UDP_RR-40 : Avg: 76231 Avg: 75419.1
UDP_RR-80 : Avg: 34578.3 Avg: 35639.1
UDP_STREAM-1 : Avg: 64684.3 Avg: 66606
UDP_STREAM-10 : Avg: 52701.2 Avg: 52959.5
UDP_STREAM-20 : Avg: 30376.4 Avg: 29704
UDP_STREAM-40 : Avg: 15685.8 Avg: 15266.5
UDP_STREAM-80 : Avg: 8415.13 Avg: 7388.97
(wins and losses)
sysbench:
NO_WA_WEIGHT WA_WEIGHT
sysbench-mysql-2 : 2135.17 per sec. 2142.51 per sec.
sysbench-mysql-5 : 4809.68 per sec. 4800.19 per sec.
sysbench-mysql-10 : 9158.59 per sec. 9157.05 per sec.
sysbench-mysql-20 : 14570.70 per sec. 14543.55 per sec.
sysbench-mysql-40 : 22130.56 per sec. 22184.82 per sec.
sysbench-mysql-80 : 20995.56 per sec. 21904.18 per sec.
sysbench-psql-2 : 1679.58 per sec. 1705.06 per sec.
sysbench-psql-5 : 3797.69 per sec. 3879.93 per sec.
sysbench-psql-10 : 7253.22 per sec. 7258.06 per sec.
sysbench-psql-20 : 11166.75 per sec. 11220.00 per sec.
sysbench-psql-40 : 17277.28 per sec. 17359.78 per sec.
sysbench-psql-80 : 17112.44 per sec. 17221.16 per sec.
(increase on the top end)
tbench:
NO_WA_WEIGHT
Throughput 685.211 MB/sec 2 clients 2 procs max_latency=0.123 ms
Throughput 1596.64 MB/sec 5 clients 5 procs max_latency=0.119 ms
Throughput 2985.47 MB/sec 10 clients 10 procs max_latency=0.262 ms
Throughput 4521.15 MB/sec 20 clients 20 procs max_latency=0.506 ms
Throughput 9438.1 MB/sec 40 clients 40 procs max_latency=2.052 ms
Throughput 8210.5 MB/sec 80 clients 80 procs max_latency=8.310 ms
WA_WEIGHT
Throughput 697.292 MB/sec 2 clients 2 procs max_latency=0.127 ms
Throughput 1596.48 MB/sec 5 clients 5 procs max_latency=0.080 ms
Throughput 2975.22 MB/sec 10 clients 10 procs max_latency=0.254 ms
Throughput 4575.14 MB/sec 20 clients 20 procs max_latency=0.502 ms
Throughput 9468.65 MB/sec 40 clients 40 procs max_latency=2.069 ms
Throughput 8631.73 MB/sec 80 clients 80 procs max_latency=8.605 ms
(increase on the top end)
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Eric reported a sysbench regression against commit:
3fed382b46 ("sched/numa: Implement NUMA node level wake_affine()")
Similarly, Rik was looking at the NAS-lu.C benchmark, which regressed
against his v3.10 enterprise kernel.
PRE (current tip/master):
ivb-ep sysbench:
2: [30 secs] transactions: 64110 (2136.94 per sec.)
5: [30 secs] transactions: 143644 (4787.99 per sec.)
10: [30 secs] transactions: 274298 (9142.93 per sec.)
20: [30 secs] transactions: 418683 (13955.45 per sec.)
40: [30 secs] transactions: 320731 (10690.15 per sec.)
80: [30 secs] transactions: 355096 (11834.28 per sec.)
hsw-ex NAS:
OMP_PROC_BIND/lu.C.x_threads_144_run_1.log: Time in seconds = 18.01
OMP_PROC_BIND/lu.C.x_threads_144_run_2.log: Time in seconds = 17.89
OMP_PROC_BIND/lu.C.x_threads_144_run_3.log: Time in seconds = 17.93
lu.C.x_threads_144_run_1.log: Time in seconds = 434.68
lu.C.x_threads_144_run_2.log: Time in seconds = 405.36
lu.C.x_threads_144_run_3.log: Time in seconds = 433.83
POST (+patch):
ivb-ep sysbench:
2: [30 secs] transactions: 64494 (2149.75 per sec.)
5: [30 secs] transactions: 145114 (4836.99 per sec.)
10: [30 secs] transactions: 278311 (9276.69 per sec.)
20: [30 secs] transactions: 437169 (14571.60 per sec.)
40: [30 secs] transactions: 669837 (22326.73 per sec.)
80: [30 secs] transactions: 631739 (21055.88 per sec.)
hsw-ex NAS:
lu.C.x_threads_144_run_1.log: Time in seconds = 23.36
lu.C.x_threads_144_run_2.log: Time in seconds = 22.96
lu.C.x_threads_144_run_3.log: Time in seconds = 22.52
This patch takes out all the shiny wake_affine() stuff and goes back to
utter basics. Between the two CPUs involved with the wakeup (the CPU
doing the wakeup and the CPU we ran on previously) pick the CPU we can
run on _now_.
This restores much of the regressions against the older kernels,
but leaves some ground in the overloaded case. The default-enabled
WA_WEIGHT (which will be introduced in the next patch) is an attempt
to address the overloaded situation.
Reported-by: Eric Farman <farman@linux.vnet.ibm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Matthew Rosato <mjrosato@linux.vnet.ibm.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: jinpuwang@gmail.com
Cc: vcaputo@pengaru.com
Fixes: 3fed382b46 ("sched/numa: Implement NUMA node level wake_affine()")
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Hackbench recently suffered a bunch of pain, first by commit:
4c77b18cf8 ("sched/fair: Make select_idle_cpu() more aggressive")
and then by commit:
c743f0a5c5 ("sched/fair, cpumask: Export for_each_cpu_wrap()")
which fixed a bug in the initial for_each_cpu_wrap() implementation
that made select_idle_cpu() even more expensive. The bug was that it
would skip over CPUs when bits were consequtive in the bitmask.
This however gave me an idea to fix select_idle_cpu(); where the old
scheme was a cliff-edge throttle on idle scanning, this introduces a
more gradual approach. Instead of stopping to scan entirely, we limit
how many CPUs we scan.
Initial benchmarks show that it mostly recovers hackbench while not
hurting anything else, except Mason's schbench, but not as bad as the
old thing.
It also appears to recover the tbench high-end, which also suffered like
hackbench.
Tested-by: Matt Fleming <matt@codeblueprint.co.uk>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Chris Mason <clm@fb.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: hpa@zytor.com
Cc: kitsunyan <kitsunyan@inbox.ru>
Cc: linux-kernel@vger.kernel.org
Cc: lvenanci@redhat.com
Cc: riel@redhat.com
Cc: xiaolong.ye@intel.com
Link: http://lkml.kernel.org/r/20170517105350.hk5m4h4jb6dfr65a@hirez.programming.kicks-ass.net
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Its an obsolete debug mechanism and future code wants to rely on
properties this undermines.
Namely, it would be good to assume that SD_OVERLAP domains have
children, but if we build the entire hierarchy with SD_OVERLAP this is
obviously false.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Now that we have no missing calls, add a warning to find multiple
calls.
By having only a single update_rq_clock() call per rq-lock section,
the section appears 'atomic' wrt time.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Kitsunyan reported desktop latency issues on his Celeron 887 because
of commit:
1b568f0aab ("sched/core: Optimize SCHED_SMT")
... even though his CPU doesn't do SMT.
The effect of running the SMT code on a !SMT part is basically a more
aggressive select_idle_cpu(). Removing the avg condition fixed things
for him.
I also know FB likes this test gone, even though other workloads like
having it.
For now, take it out by default, until we get a better idea.
Reported-by: kitsunyan <kitsunyan@inbox.ru>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Chris Mason <clm@fb.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Mike Galbraith <umgwanakikbuti@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Bring arch_scale_cpu_capacity() in line with the recent change of its
arch_scale_freq_capacity() sibling in commit dfbca41f34 ("sched:
Optimize freq invariant accounting") from weak function to #define to
allow inlining of the function.
While at it, remove the ARCH_CAPACITY sched_feature as well. With the
change to #define there isn't a straightforward way to allow runtime
switch between an arch implementation and the default implementation of
arch_scale_cpu_capacity() using sched_feature. The default was to use
the arch-specific implementation, but only the arm architecture provides
one and that is essentially equivalent to the default implementation.
Signed-off-by: Morten Rasmussen <morten.rasmussen@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Dietmar Eggemann <Dietmar.Eggemann@arm.com>
Cc: Juri Lelli <Juri.Lelli@arm.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: daniel.lezcano@linaro.org
Cc: mturquette@baylibre.com
Cc: pang.xunlei@zte.com.cn
Cc: rjw@rjwysocki.net
Cc: sgurrappadi@nvidia.com
Cc: vincent.guittot@linaro.org
Cc: yuyang.du@intel.com
Link: http://lkml.kernel.org/r/1439569394-11974-3-git-send-email-morten.rasmussen@arm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Variable sched_numa_balancing is available for both CONFIG_SCHED_DEBUG
and !CONFIG_SCHED_DEBUG. All code paths now check for
sched_numa_balancing. Hence remove sched_feat(NUMA).
Suggested-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/1439290813-6683-4-git-send-email-srikar@linux.vnet.ibm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
In case there are problems with the aging on attach, provide a debug
knob to turn it off.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Byungchul Park <byungchul.park@lge.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-kernel@vger.kernel.org
Cc: yuyang.du@intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The current load balancer may not try to prevent a task from moving
out of a preferred node to a less preferred node. The reason for this
being:
- Since sched features NUMA and NUMA_RESIST_LOWER are disabled by
default, migrate_degrades_locality() always returns false.
- Even if NUMA_RESIST_LOWER were to be enabled, if its cache hot,
migrate_degrades_locality() never gets called.
The above behaviour can mean that tasks can move out of their
preferred node but they may be eventually be brought back to their
preferred node by numa balancer (due to higher numa faults).
To avoid the above, this commit merges migrate_degrades_locality() and
migrate_improves_locality(). It also replaces 3 sched features NUMA,
NUMA_FAVOUR_HIGHER and NUMA_RESIST_LOWER by a single sched feature
NUMA.
Signed-off-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Mike Galbraith <efault@gmx.de>
Link: http://lkml.kernel.org/r/1434455762-30857-2-git-send-email-srikar@linux.vnet.ibm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
When debugging the latencies on a 40 core box, where we hit 300 to
500 microsecond latencies, I found there was a huge contention on the
runqueue locks.
Investigating it further, running ftrace, I found that it was due to
the pulling of RT tasks.
The test that was run was the following:
cyclictest --numa -p95 -m -d0 -i100
This created a thread on each CPU, that would set its wakeup in iterations
of 100 microseconds. The -d0 means that all the threads had the same
interval (100us). Each thread sleeps for 100us and wakes up and measures
its latencies.
cyclictest is maintained at:
git://git.kernel.org/pub/scm/linux/kernel/git/clrkwllms/rt-tests.git
What happened was another RT task would be scheduled on one of the CPUs
that was running our test, when the other CPU tests went to sleep and
scheduled idle. This caused the "pull" operation to execute on all
these CPUs. Each one of these saw the RT task that was overloaded on
the CPU of the test that was still running, and each one tried
to grab that task in a thundering herd way.
To grab the task, each thread would do a double rq lock grab, grabbing
its own lock as well as the rq of the overloaded CPU. As the sched
domains on this box was rather flat for its size, I saw up to 12 CPUs
block on this lock at once. This caused a ripple affect with the
rq locks especially since the taking was done via a double rq lock, which
means that several of the CPUs had their own rq locks held while trying
to take this rq lock. As these locks were blocked, any wakeups or load
balanceing on these CPUs would also block on these locks, and the wait
time escalated.
I've tried various methods to lessen the load, but things like an
atomic counter to only let one CPU grab the task wont work, because
the task may have a limited affinity, and we may pick the wrong
CPU to take that lock and do the pull, to only find out that the
CPU we picked isn't in the task's affinity.
Instead of doing the PULL, I now have the CPUs that want the pull to
send over an IPI to the overloaded CPU, and let that CPU pick what
CPU to push the task to. No more need to grab the rq lock, and the
push/pull algorithm still works fine.
With this patch, the latency dropped to just 150us over a 20 hour run.
Without the patch, the huge latencies would trigger in seconds.
I've created a new sched feature called RT_PUSH_IPI, which is enabled
by default.
When RT_PUSH_IPI is not enabled, the old method of grabbing the rq locks
and having the pulling CPU do the work is implemented. When RT_PUSH_IPI
is enabled, the IPI is sent to the overloaded CPU to do a push.
To enabled or disable this at run time:
# mount -t debugfs nodev /sys/kernel/debug
# echo RT_PUSH_IPI > /sys/kernel/debug/sched_features
or
# echo NO_RT_PUSH_IPI > /sys/kernel/debug/sched_features
Update: This original patch would send an IPI to all CPUs in the RT overload
list. But that could theoretically cause the reverse issue. That is, there
could be lots of overloaded RT queues and one CPU lowers its priority. It would
then send an IPI to all the overloaded RT queues and they could then all try
to grab the rq lock of the CPU lowering its priority, and then we have the
same problem.
The latest design sends out only one IPI to the first overloaded CPU. It tries to
push any tasks that it can, and then looks for the next overloaded CPU that can
push to the source CPU. The IPIs stop when all overloaded CPUs that have pushable
tasks that have priorities greater than the source CPU are covered. In case the
source CPU lowers its priority again, a flag is set to tell the IPI traversal to
restart with the first RT overloaded CPU after the source CPU.
Parts-suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Joern Engel <joern@purestorage.com>
Cc: Clark Williams <williams@redhat.com>
Cc: Mike Galbraith <umgwanakikbuti@gmail.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20150318144946.2f3cc982@gandalf.local.home
Signed-off-by: Ingo Molnar <mingo@kernel.org>
It is better not to think about compute capacity as being equivalent
to "CPU power". The upcoming "power aware" scheduler work may create
confusion with the notion of energy consumption if "power" is used too
liberally.
Let's rename the following feature flags since they do relate to capacity:
SD_SHARE_CPUPOWER -> SD_SHARE_CPUCAPACITY
ARCH_POWER -> ARCH_CAPACITY
NONTASK_POWER -> NONTASK_CAPACITY
Signed-off-by: Nicolas Pitre <nico@linaro.org>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Cc: Vincent Guittot <vincent.guittot@linaro.org>
Cc: Daniel Lezcano <daniel.lezcano@linaro.org>
Cc: Morten Rasmussen <morten.rasmussen@arm.com>
Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>
Cc: linaro-kernel@lists.linaro.org
Cc: Andy Fleming <afleming@freescale.com>
Cc: Anton Blanchard <anton@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Grant Likely <grant.likely@linaro.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Paul Gortmaker <paul.gortmaker@windriver.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Preeti U Murthy <preeti@linux.vnet.ibm.com>
Cc: Rob Herring <robh+dt@kernel.org>
Cc: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Cc: Toshi Kani <toshi.kani@hp.com>
Cc: Vasant Hegde <hegdevasant@linux.vnet.ibm.com>
Cc: Vincent Guittot <vincent.guittot@linaro.org>
Cc: devicetree@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Cc: linuxppc-dev@lists.ozlabs.org
Link: http://lkml.kernel.org/n/tip-e93lpnxb87owfievqatey6b5@git.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Just as "sched: Favour moving tasks towards the preferred node" favours
moving tasks towards nodes with a higher number of recorded NUMA hinting
faults, this patch resists moving tasks towards nodes with lower faults.
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1381141781-10992-24-git-send-email-mgorman@suse.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>
This patch favours moving tasks towards NUMA node that recorded a higher
number of NUMA faults during active load balancing. Ideally this is
self-reinforcing as the longer the task runs on that node, the more faults
it should incur causing task_numa_placement to keep the task running on that
node. In reality a big weakness is that the nodes CPUs can be overloaded
and it would be more efficient to queue tasks on an idle node and migrate
to the new node. This would require additional smarts in the balancer so
for now the balancer will simply prefer to place the task on the preferred
node for a PTE scans which is controlled by the numa_balancing_settle_count
sysctl. Once the settle_count number of scans has complete the schedule
is free to place the task on an alternative node if the load is imbalanced.
[srikar@linux.vnet.ibm.com: Fixed statistics]
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
[ Tunable and use higher faults instead of preferred. ]
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1381141781-10992-23-git-send-email-mgorman@suse.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>
PTE scanning and NUMA hinting fault handling is expensive so commit
5bca2303 ("mm: sched: numa: Delay PTE scanning until a task is scheduled
on a new node") deferred the PTE scan until a task had been scheduled on
another node. The problem is that in the purely shared memory case that
this may never happen and no NUMA hinting fault information will be
captured. We are not ruling out the possibility that something better
can be done here but for now, this patch needs to be reverted and depend
entirely on the scan_delay to avoid punishing short-lived processes.
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1381141781-10992-16-git-send-email-mgorman@suse.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>
As mentioned by Ingo, the SCHED_FEAT_OWNER_SPIN scheduler
feature bit was really just an early hack to make with/without
mutex-spinning testable. So it is no longer necessary.
This patch removes the SCHED_FEAT_OWNER_SPIN feature bit and
move the mutex spinning code from kernel/sched/core.c back to
kernel/mutex.c which is where they should belong.
Signed-off-by: Waiman Long <Waiman.Long@hp.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Chandramouleeswaran Aswin <aswin@hp.com>
Cc: Davidlohr Bueso <davidlohr.bueso@hp.com>
Cc: Norton Scott J <scott.norton@hp.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: David Howells <dhowells@redhat.com>
Cc: Dave Jones <davej@redhat.com>
Cc: Clark Williams <williams@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1366226594-5506-2-git-send-email-Waiman.Long@hp.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.18 (GNU/Linux)
iQIcBAABAgAGBQJQx0kQAAoJEHzG/DNEskfi4fQP/R5PRovayroZALBMLnVJDaLD
Ttr9p40VNXbiJ+MfRgatJjSSJZ4Jl+fC3NEqBhcwVZhckZZb9R2s0WtrSQo5+ZbB
vdRfiuKoCaKM4cSZ08C12uTvsF6xjhjd27CTUlMkyOcDoKxMEFKelv0hocSxe4Wo
xqlv3eF+VsY7kE1BNbgBP06SX4tDpIHRxXfqJPMHaSKQmre+cU0xG2GcEu3QGbHT
DEDTI788YSaWLmBfMC+kWoaQl1+bV/FYvavIAS8/o4K9IKvgR42VzrXmaFaqrbgb
72ksa6xfAi57yTmZHqyGmts06qYeBbPpKI+yIhCMInxA9CY3lPbvHppRf0RQOyzj
YOi4hovGEMJKE+BCILukhJcZ9jCTtS3zut6v1rdvR88f4y7uhR9RfmRfsxuW7PNj
3Rmh191+n0lVWDmhOs2psXuCLJr3LEiA0dFffN1z8REUTtTAZMsj8Rz+SvBNAZDR
hsJhERVeXB6X5uQ5rkLDzbn1Zic60LjVw7LIp6SF2OYf/YKaF8vhyWOA8dyCEu8W
CGo7AoG0BO8tIIr8+LvFe8CweypysZImx4AjCfIs4u9pu/v11zmBvO9NO5yfuObF
BreEERYgTes/UITxn1qdIW4/q+Nr0iKO3CTqsmu6L1GfCz3/XzPGs3U26fUhllqi
Ka0JKgnWvsa6ez6FSzKI
=ivQa
-----END PGP SIGNATURE-----
Merge tag 'balancenuma-v11' of git://git.kernel.org/pub/scm/linux/kernel/git/mel/linux-balancenuma
Pull Automatic NUMA Balancing bare-bones from Mel Gorman:
"There are three implementations for NUMA balancing, this tree
(balancenuma), numacore which has been developed in tip/master and
autonuma which is in aa.git.
In almost all respects balancenuma is the dumbest of the three because
its main impact is on the VM side with no attempt to be smart about
scheduling. In the interest of getting the ball rolling, it would be
desirable to see this much merged for 3.8 with the view to building
scheduler smarts on top and adapting the VM where required for 3.9.
The most recent set of comparisons available from different people are
mel: https://lkml.org/lkml/2012/12/9/108
mingo: https://lkml.org/lkml/2012/12/7/331
tglx: https://lkml.org/lkml/2012/12/10/437
srikar: https://lkml.org/lkml/2012/12/10/397
The results are a mixed bag. In my own tests, balancenuma does
reasonably well. It's dumb as rocks and does not regress against
mainline. On the other hand, Ingo's tests shows that balancenuma is
incapable of converging for this workloads driven by perf which is bad
but is potentially explained by the lack of scheduler smarts. Thomas'
results show balancenuma improves on mainline but falls far short of
numacore or autonuma. Srikar's results indicate we all suffer on a
large machine with imbalanced node sizes.
My own testing showed that recent numacore results have improved
dramatically, particularly in the last week but not universally.
We've butted heads heavily on system CPU usage and high levels of
migration even when it shows that overall performance is better.
There are also cases where it regresses. Of interest is that for
specjbb in some configurations it will regress for lower numbers of
warehouses and show gains for higher numbers which is not reported by
the tool by default and sometimes missed in treports. Recently I
reported for numacore that the JVM was crashing with
NullPointerExceptions but currently it's unclear what the source of
this problem is. Initially I thought it was in how numacore batch
handles PTEs but I'm no longer think this is the case. It's possible
numacore is just able to trigger it due to higher rates of migration.
These reports were quite late in the cycle so I/we would like to start
with this tree as it contains much of the code we can agree on and has
not changed significantly over the last 2-3 weeks."
* tag 'balancenuma-v11' of git://git.kernel.org/pub/scm/linux/kernel/git/mel/linux-balancenuma: (50 commits)
mm/rmap, migration: Make rmap_walk_anon() and try_to_unmap_anon() more scalable
mm/rmap: Convert the struct anon_vma::mutex to an rwsem
mm: migrate: Account a transhuge page properly when rate limiting
mm: numa: Account for failed allocations and isolations as migration failures
mm: numa: Add THP migration for the NUMA working set scanning fault case build fix
mm: numa: Add THP migration for the NUMA working set scanning fault case.
mm: sched: numa: Delay PTE scanning until a task is scheduled on a new node
mm: sched: numa: Control enabling and disabling of NUMA balancing if !SCHED_DEBUG
mm: sched: numa: Control enabling and disabling of NUMA balancing
mm: sched: Adapt the scanning rate if a NUMA hinting fault does not migrate
mm: numa: Use a two-stage filter to restrict pages being migrated for unlikely task<->node relationships
mm: numa: migrate: Set last_nid on newly allocated page
mm: numa: split_huge_page: Transfer last_nid on tail page
mm: numa: Introduce last_nid to the page frame
sched: numa: Slowly increase the scanning period as NUMA faults are handled
mm: numa: Rate limit setting of pte_numa if node is saturated
mm: numa: Rate limit the amount of memory that is migrated between nodes
mm: numa: Structures for Migrate On Fault per NUMA migration rate limiting
mm: numa: Migrate pages handled during a pmd_numa hinting fault
mm: numa: Migrate on reference policy
...
Due to the fact that migrations are driven by the CPU a task is running
on there is no point tracking NUMA faults until one task runs on a new
node. This patch tracks the first node used by an address space. Until
it changes, PTE scanning is disabled and no NUMA hinting faults are
trapped. This should help workloads that are short-lived, do not care
about NUMA placement or have bound themselves to a single node.
This takes advantage of the logic in "mm: sched: numa: Implement slow
start for working set sampling" to delay when the checks are made. This
will take advantage of processes that set their CPU and node bindings
early in their lifetime. It will also potentially allow any initial load
balancing to take place.
Signed-off-by: Mel Gorman <mgorman@suse.de>
This patch adds Kconfig options and kernel parameters to allow the
enabling and disabling of automatic NUMA balancing. The existance
of such a switch was and is very important when debugging problems
related to transparent hugepages and we should have the same for
automatic NUMA placement.
Signed-off-by: Mel Gorman <mgorman@suse.de>
NOTE: This patch is based on "sched, numa, mm: Add fault driven
placement and migration policy" but as it throws away all the policy
to just leave a basic foundation I had to drop the signed-offs-by.
This patch creates a bare-bones method for setting PTEs pte_numa in the
context of the scheduler that when faulted later will be faulted onto the
node the CPU is running on. In itself this does nothing useful but any
placement policy will fundamentally depend on receiving hints on placement
from fault context and doing something intelligent about it.
Signed-off-by: Mel Gorman <mgorman@suse.de>
Acked-by: Rik van Riel <riel@redhat.com>
As per the recent discussion with Mike and Linus, make it easier to
test with/without this feature. No change in default behavior.
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Link: http://lkml.kernel.org/n/tip-izoxq4haeg4mTognnDbwcevt@git.kernel.org
Heteregeneous ARM platform uses arch_scale_freq_power function
to reflect the relative capacity of each core
Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1341826026-6504-6-git-send-email-vincent.guittot@linaro.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Commit beac4c7e4a ("sched: Remove AFFINE_WAKEUPS feature") removed
use of the flag but left the definition. Get rid of it.
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Link: http://lkml.kernel.org/r/1345090865-20851-1-git-send-email-namhyung@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Commits 367456c756 ("sched: Ditch per cgroup task lists for
load-balancing") and 5d6523ebd ("sched: Fix load-balance wreckage")
left some more wreckage.
By setting loop_max unconditionally to ->nr_running load-balancing
could take a lot of time on very long runqueues (hackbench!). So keep
the sysctl as max limit of the amount of tasks we'll iterate.
Furthermore, the min load filter for migration completely fails with
cgroups since inequality in per-cpu state can easily lead to such
small loads :/
Furthermore the change to add new tasks to the tail of the queue
instead of the head seems to have some effect.. not quite sure I
understand why.
Combined these fixes solve the huge hackbench regression reported by
Tim when hackbench is ran in a cgroup.
Reported-by: Tim Chen <tim.c.chen@linux.intel.com>
Acked-by: Tim Chen <tim.c.chen@linux.intel.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/r/1335365763.28150.267.camel@twins
[ got rid of the CONFIG_PREEMPT tuning and made small readability edits ]
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Now that we initialize jump_labels before sched_init() we can use them
for the debug features without having to worry about a window where
they have the wrong setting.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/n/tip-vpreo4hal9e0kzqmg5y0io2k@git.kernel.org
Signed-off-by: Ingo Molnar <mingo@elte.hu>
There's too many sched*.[ch] files in kernel/, give them their own
directory.
(No code changed, other than Makefile glue added.)
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Chen Yu