Skip to contentSkip to conversation

fc7f0dd3817: -2.1% will-it-scale.per_thread_ops

LKP Huang Ying 2 participants 3 messages Jan 22, 2015 – Jan 27, 2015

lore
Share & tools
lore

Huang Ying

author
#0 lore raw

FYI, we noticed the below changes on

commit fc7f0dd381720ea5ee5818645f7d0e9dece41cb0 ("kernel: avoid overflow in cmp_range")

testbox/testcase/testparams: lituya/will-it-scale/powersave-mmap2

7ad4b4ae5757b896 fc7f0dd381720ea5ee5818645f
---------------- --------------------------

         %stddev     %change         %stddev
             \          |                \  
    252693 ±  0%      -2.2%     247031 ±  0%  will-it-scale.per_thread_ops
      0.18 ±  0%      +1.8%       0.19 ±  0%  will-it-scale.scalability
     43536 ± 24%    +276.2%     163774 ± 33%  sched_debug.cpu#6.ttwu_local
      3.55 ±  2%     +36.2%       4.84 ±  2%  perf-profile.cpu-cycles.___might_sleep.unmap_page_range.unmap_single_vma.unmap_vmas.unmap_region
      8.49 ± 12%     -29.5%       5.99 ±  5%  perf-profile.cpu-cycles._raw_spin_lock_irqsave.__percpu_counter_add.do_munmap.vm_munmap.sys_munmap
     12.27 ±  8%     -20.2%       9.80 ±  3%  perf-profile.cpu-cycles.__percpu_counter_add.do_munmap.vm_munmap.sys_munmap.system_call_fastpath
      7.45 ±  7%     -20.8%       5.90 ±  5%  perf-profile.cpu-cycles._raw_spin_lock_irqsave.__percpu_counter_add.__vm_enough_memory.selinux_vm_enough_memory.security_vm_enough_memory_mm
     11.11 ±  3%     -12.9%       9.67 ±  3%  perf-profile.cpu-cycles.__percpu_counter_add.__vm_enough_memory.selinux_vm_enough_memory.security_vm_enough_memory_mm.mmap_region
      2.46 ±  3%     +13.1%       2.78 ±  2%  perf-profile.cpu-cycles.___might_sleep.unmap_single_vma.unmap_vmas.unmap_region.do_munmap
     11.42 ±  3%     -12.3%      10.01 ±  2%  perf-profile.cpu-cycles.__vm_enough_memory.selinux_vm_enough_memory.security_vm_enough_memory_mm.mmap_region.do_mmap_pgoff
     12.39 ±  3%     -11.2%      11.00 ±  2%  perf-profile.cpu-cycles.selinux_vm_enough_memory.security_vm_enough_memory_mm.mmap_region.do_mmap_pgoff.vm_mmap_pgoff
     12.45 ±  3%     -11.1%      11.07 ±  2%  perf-profile.cpu-cycles.security_vm_enough_memory_mm.mmap_region.do_mmap_pgoff.vm_mmap_pgoff.sys_mmap_pgoff
     14.38 ±  1%      +9.5%      15.75 ±  1%  perf-profile.cpu-cycles.unmap_page_range.unmap_single_vma.unmap_vmas.unmap_region.do_munmap

testbox/testcase/testparams: lituya/will-it-scale/performance-mmap2

7ad4b4ae5757b896 fc7f0dd381720ea5ee5818645f
---------------- --------------------------

    268761 ±  0%      -2.1%     263177 ±  0%  will-it-scale.per_thread_ops
      0.18 ±  0%      +1.8%       0.19 ±  0%  will-it-scale.scalability
      0.01 ± 37%     -99.3%       0.00 ± 12%  sched_debug.rt_rq[10]:/.rt_time
    104123 ± 41%     -63.7%      37788 ± 45%  sched_debug.cpu#5.ttwu_local
    459901 ± 48%     +60.7%     739071 ± 31%  sched_debug.cpu#6.ttwu_count
   1858053 ± 12%     -36.9%    1171826 ± 38%  sched_debug.cpu#10.sched_goidle
   3716823 ± 12%     -36.9%    2344353 ± 38%  sched_debug.cpu#10.nr_switches
   3777468 ± 11%     -36.9%    2383575 ± 36%  sched_debug.cpu#10.sched_count
        36 ± 28%     -40.9%         21 ±  7%  sched_debug.cpu#6.cpu_load[1]
     18042 ± 17%     +54.0%      27789 ± 30%  sched_debug.cfs_rq[4]:/.exec_clock
        56 ± 17%     -48.8%         29 ±  5%  sched_debug.cfs_rq[6]:/.runnable_load_avg
        36 ± 29%     +43.6%         52 ± 11%  sched_debug.cpu#4.load
    594415 ±  4%     +82.4%    1084432 ± 18%  sched_debug.cpu#2.ttwu_count
        15 ±  0%     +51.1%         22 ± 14%  sched_debug.cpu#4.cpu_load[4]
      2077 ± 11%     -36.7%       1315 ± 15%  sched_debug.cpu#6.curr->pid
        11 ± 28%     +48.6%         17 ± 23%  sched_debug.cpu#7.cpu_load[4]
      0.00 ± 20%     +77.0%       0.00 ± 26%  sched_debug.rt_rq[5]:/.rt_time
        16 ±  5%     +52.1%         24 ±  9%  sched_debug.cpu#4.cpu_load[3]
        17 ± 11%     +50.0%         26 ±  8%  sched_debug.cpu#4.cpu_load[2]
     48035 ±  7%     -22.2%      37362 ± 24%  sched_debug.cfs_rq[12]:/.exec_clock
        34 ± 12%     -24.5%         25 ± 20%  sched_debug.cfs_rq[12]:/.runnable_load_avg
        33 ± 11%     -24.2%         25 ± 20%  sched_debug.cpu#12.cpu_load[4]
        19 ± 25%     +50.9%         28 ±  3%  sched_debug.cpu#4.cpu_load[1]
        66 ± 17%     -24.7%         49 ±  5%  sched_debug.cpu#6.load
    421462 ± 16%     +18.8%     500676 ± 13%  sched_debug.cfs_rq[1]:/.min_vruntime
      3.60 ±  0%     +35.4%       4.87 ±  0%  perf-profile.cpu-cycles.___might_sleep.unmap_page_range.unmap_single_vma.unmap_vmas.unmap_region
        44 ±  9%     +37.9%         60 ± 17%  sched_debug.cpu#3.load
        37 ±  6%     -17.9%         30 ± 15%  sched_debug.cpu#15.cpu_load[3]
      6.96 ±  4%     -10.4%       6.24 ±  3%  perf-profile.cpu-cycles._raw_spin_lock_irqsave.__percpu_counter_add.do_munmap.vm_munmap.sys_munmap
        36 ±  6%     +24.1%         44 ±  2%  sched_debug.cpu#2.load
        39 ±  7%     -16.9%         32 ± 12%  sched_debug.cpu#15.cpu_load[2]
   1528695 ±  6%     -19.5%    1230190 ± 16%  sched_debug.cpu#10.ttwu_count
        36 ±  6%     +27.3%         46 ±  9%  sched_debug.cpu#10.load
       447 ±  3%     -13.9%        385 ± 10%  sched_debug.cfs_rq[15]:/.tg_runnable_contrib
     20528 ±  3%     -13.8%      17701 ± 10%  sched_debug.cfs_rq[15]:/.avg->runnable_avg_sum
    634808 ±  6%     +50.3%     954347 ± 24%  sched_debug.cpu#2.sched_goidle
   1270648 ±  6%     +50.3%    1909528 ± 24%  sched_debug.cpu#2.nr_switches
   1284042 ±  6%     +51.4%    1944604 ± 23%  sched_debug.cpu#2.sched_count
        55 ± 11%     +28.7%         71 ±  4%  sched_debug.cpu#8.cpu_load[0]
      6.39 ±  0%      -8.7%       5.84 ±  2%  perf-profile.cpu-cycles._raw_spin_lock_irqsave.__percpu_counter_add.__vm_enough_memory.selinux_vm_enough_memory.security_vm_enough_memory_mm
     48721 ± 11%     +19.1%      58037 ±  5%  sched_debug.cpu#11.nr_load_updates
        53 ±  9%     +16.1%         62 ±  1%  sched_debug.cpu#8.cpu_load[1]
      1909 ±  0%     +22.2%       2333 ±  9%  sched_debug.cpu#3.curr->pid
      0.95 ±  4%      -8.4%       0.87 ±  4%  perf-profile.cpu-cycles.file_map_prot_check.selinux_mmap_file.security_mmap_file.vm_mmap_pgoff.sys_mmap_pgoff
    567608 ±  8%     +11.0%     629780 ±  4%  sched_debug.cfs_rq[14]:/.min_vruntime
    804637 ± 15%     +24.4%    1000664 ± 13%  sched_debug.cpu#3.ttwu_count
    684460 ±  5%      -9.6%     618867 ±  3%  sched_debug.cpu#14.avg_idle
      1.02 ±  4%      -7.2%       0.94 ±  4%  perf-profile.cpu-cycles.selinux_mmap_file.security_mmap_file.vm_mmap_pgoff.sys_mmap_pgoff.sys_mmap
      2605 ±  2%      -5.8%       2454 ±  5%  slabinfo.kmalloc-96.active_objs
      2605 ±  2%      -5.8%       2454 ±  5%  slabinfo.kmalloc-96.num_objs
        50 ±  4%     +11.3%         56 ±  1%  sched_debug.cfs_rq[8]:/.runnable_load_avg
      1.15 ±  4%      -6.4%       1.08 ±  4%  perf-profile.cpu-cycles.security_mmap_file.vm_mmap_pgoff.sys_mmap_pgoff.sys_mmap.system_call_fastpath
      1.07 ±  2%      +9.7%       1.17 ±  3%  perf-profile.cpu-cycles.vma_compute_subtree_gap.__vma_link_rb.vma_link.mmap_region.do_mmap_pgoff

To reproduce:

apt-get install ruby ruby-oj
git clone git://git.kernel.org/pub/scm/linux/kernel/git/wfg/lkp-tests.git
cd lkp-tests
bin/setup-local job.yaml # the job file attached in this email
bin/run-local job.yaml

Disclaimer:
Results have been estimated based on internal Intel analysis and are provided for informational purposes only. Any difference in system hardware or software design or configuration may affect actual performance.

Thanks,
Huang, Ying

_______________________________________________
LKP mailing list
LKP(a)linux.intel.com

---
testcase: will-it-scale
default-monitors:
wait: pre-test
uptime:
iostat:
vmstat:
numa-numastat:
numa-vmstat:
numa-meminfo:
proc-vmstat:
proc-stat:
meminfo:
slabinfo:
interrupts:
lock_stat:
latency_stats:
softirqs:
bdi_dev_mapping:
diskstats:
nfsstat:
cpuidle:
cpufreq-stats:
turbostat:
pmeter:
sched_debug:
interval: 10
default_watchdogs:
watch-oom:
watchdog:
cpufreq_governor:
- powersave
commit: ec6f34e5b552fb0a52e6aae1a5afbbb1605cc6cc
model: Grantley Haswell
nr_cpu: 16
memory: 16G
hdd_partitions:
swap_partitions:
rootfs_partition:
perf-profile:
freq: 800
will-it-scale:
test:
- mmap2
testbox: lituya
tbox_group: lituya
kconfig: x86_64-rhel
enqueue_time: 2015-01-18 14:10:07.541442957 +08:00
head_commit: b213d55915f2ee6748ba62f743b5e70564ab31e7
base_commit: ec6f34e5b552fb0a52e6aae1a5afbbb1605cc6cc
branch: linux-devel/devel-hourly-2015011917
kernel: "/kernel/x86_64-rhel/ec6f34e5b552fb0a52e6aae1a5afbbb1605cc6cc/vmlinuz-3.19.0-rc5-gec6f34e"
user: lkp
queue: cyclic
rootfs: debian-x86_64-2015-01-13.cgz
result_root: "/result/lituya/will-it-scale/powersave-mmap2/debian-x86_64-2015-01-13.cgz/x86_64-rhel/ec6f34e5b552fb0a52e6aae1a5afbbb1605cc6cc/0"
job_file: "/lkp/scheduled/lituya/cyclic_will-it-scale-powersave-mmap2-x86_64-rhel-BASE-ec6f34e5b552fb0a52e6aae1a5afbbb1605cc6cc-0.yaml"
dequeue_time: 2015-01-19 18:08:35.232473498 +08:00
job_state: finished
loadavg: 11.32 6.60 2.70 1/178 7099
start_time: '1421662149'
end_time: '1421662453'
version: "/lkp/lkp/.src-20150119-113749"

./runtest.py mmap2 32 1 8 12 16

#1 lore raw

Hi Huang (and others on the lkp(a)01.org mailing list),

Not sure how to interpret the numbers you provided. Is there a web page that you would suggest looking at that I might read up? It wouldn't surprise me if the changes made resulted in slightly decreased performance - the number of commands shown in disassembly of the code not surprisingly increases.

Thank you.

Lou

Quoted 24 lines · Huang Ying

FYI, we noticed the below changes on

commit fc7f0dd381720ea5ee5818645f7d0e9dece41cb0 ("kernel: avoid overflow in cmp_range")

testbox/testcase/testparams: lituya/will-it-scale/powersave-mmap2

7ad4b4ae5757b896 fc7f0dd381720ea5ee5818645f
---------------- --------------------------
%stddev %change %stddev
\ | \ ...
Disclaimer:
Results have been estimated based on internal Intel analysis and are provided for informational purposes only. Any difference in system hardware or software design or configuration may affect actual performance.

Thanks,
Huang, Ying

<job.yaml><reproduce.txt>_______________________________________________ LKP mailing list
LKP(a)linux.intel.com

author
#2 lore raw
Quoted 4 lines · Louis Langholtz

Hi Huang (and others on the lkp(a)01.org mailing list),

Not sure how to interpret the numbers you provided. Is there a web page that you would suggest looking at that I might read up?

Sorry we have no web page. Most words in the email is straightforward.

testbox/testcase/testparams: lituya/will-it-scale/powersave-mmap2

test machine: lituya, it is a Haswell-EP machine
test case: will-it-scale (a benchmark to measure performance via invoking some system call in a loop)
test params: powersave-mmap2: powersave CPU frequency governor is used, the sub test case in will-it-scale is: mmap2

7ad4b4ae5757b896 fc7f0dd381720ea5ee5818645f

Here is the parent commit ID and the commit ID for your patch.

---------------- --------------------------

         %stddev     %change         %stddev
             \          |                \  
    252693 ±  0%      -2.2%     247031 ±  0%  will-it-scale.per_thread_ops

252693 and 247031 is the score from the benchmark 0% is the standard deviation for the score -2.2% is the change between the parent and your patch.

Quoted 3 lines

It wouldn't surprise me if the changes made resulted in slightly decreased performance - the number of commands shown in disassembly of the code not surprisingly increases.

Best Regards,
Huang, Ying

Quoted 27 lines · Huang Ying
Quoted 25 lines

FYI, we noticed the below changes on

commit fc7f0dd381720ea5ee5818645f7d0e9dece41cb0 ("kernel: avoid overflow in cmp_range")

testbox/testcase/testparams: lituya/will-it-scale/powersave-mmap2

7ad4b4ae5757b896 fc7f0dd381720ea5ee5818645f
---------------- --------------------------
%stddev %change %stddev
\ | \ ...
Disclaimer:
Results have been estimated based on internal Intel analysis and are provided for informational purposes only. Any difference in system hardware or software design or configuration may affect actual performance.

Thanks,
Huang, Ying

<job.yaml><reproduce.txt>_______________________________________________ LKP mailing list
LKP(a)linux.intel.com