Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]
Groups > linux.kernel > #1409991 > unrolled thread
| Started by | "Huang\, Ying" <ying.huang@intel.com> |
|---|---|
| First post | 2016-05-31 10:40 +0200 |
| Last post | 2016-06-02 02:50 +0200 |
| Articles | 8 — 3 participants |
Back to article view | Back to linux.kernel
This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by
below is the oldest one visible, not the original post.
Re: [LKP] [lkp] [sched/fair] 53d3bc773e: hackbench.throughput -32.9% regression "Huang\, Ying" <ying.huang@intel.com> - 2016-05-31 10:40 +0200
Re: [LKP] [lkp] [sched/fair] 53d3bc773e: hackbench.throughput -32.9% regression Peter Zijlstra <peterz@infradead.org> - 2016-05-31 14:50 +0200
Re: [LKP] [lkp] [sched/fair] 53d3bc773e: hackbench.throughput -32.9% regression "Huang\, Ying" <ying.huang@intel.com> - 2016-06-01 07:10 +0200
Re: [LKP] [lkp] [sched/fair] 53d3bc773e: hackbench.throughput -32.9% regression Peter Zijlstra <peterz@infradead.org> - 2016-06-01 10:50 +0200
Re: [LKP] [lkp] [sched/fair] 53d3bc773e: hackbench.throughput -32.9% regression "Huang\, Ying" <ying.huang@intel.com> - 2016-06-01 11:00 +0200
Re: [LKP] [lkp] [sched/fair] 53d3bc773e: hackbench.throughput -32.9% regression Mike Galbraith <umgwanakikbuti@gmail.com> - 2016-06-01 11:50 +0200
Re: [LKP] [lkp] [sched/fair] 53d3bc773e: hackbench.throughput -32.9% regression "Huang\, Ying" <ying.huang@intel.com> - 2016-06-02 02:30 +0200
Re: [LKP] [lkp] [sched/fair] 53d3bc773e: hackbench.throughput -32.9% regression Mike Galbraith <umgwanakikbuti@gmail.com> - 2016-06-02 02:50 +0200
| From | "Huang\, Ying" <ying.huang@intel.com> |
|---|---|
| Date | 2016-05-31 10:40 +0200 |
| Subject | Re: [LKP] [lkp] [sched/fair] 53d3bc773e: hackbench.throughput -32.9% regression |
| Message-ID | <rENnr-iy-21@gated-at.bofh.it> |
Hi, Ingo,
Part of the regression has been recovered in v4.7-rc1 from -32.9% to
-9.8%. But there is still some regression. Is it possible for fully
restore it?
Details are as below.
Best Regards,
Huang, Ying
=========================================================================================
compiler/cpufreq_governor/ipc/kconfig/mode/nr_threads/rootfs/tbox_group/testcase:
gcc-4.9/performance/socket/x86_64-rhel/threads/50%/debian-x86_64-2015-02-07.cgz/ivb42/hackbench
commit:
c5114626f33b62fa7595e57d87f33d9d1f8298a2
53d3bc773eaa7ab1cf63585e76af7ee869d5e709
v4.7-rc1
c5114626f33b62fa 53d3bc773eaa7ab1cf63585e76 v4.7-rc1
---------------- -------------------------- --------------------------
%stddev %change %stddev %change %stddev
\ | \ | \
196590 ± 0% -32.9% 131963 ± 2% -9.8% 177231 ± 0% hackbench.throughput
602.66 ± 0% +2.8% 619.27 ± 2% +0.3% 604.66 ± 0% hackbench.time.elapsed_time
602.66 ± 0% +2.8% 619.27 ± 2% +0.3% 604.66 ± 0% hackbench.time.elapsed_time.max
1.76e+08 ± 3% +236.0% 5.914e+08 ± 2% -49.6% 88783232 ± 5% hackbench.time.involuntary_context_switches
208664 ± 2% +26.0% 262929 ± 3% +15.7% 241377 ± 0% hackbench.time.minor_page_faults
4401 ± 0% +5.7% 4650 ± 0% -8.1% 4043 ± 0% hackbench.time.percent_of_cpu_this_job_got
25256 ± 0% +10.2% 27842 ± 2% -7.7% 23311 ± 0% hackbench.time.system_time
1272 ± 0% -24.5% 961.37 ± 2% -10.4% 1140 ± 0% hackbench.time.user_time
7.64e+08 ± 1% +131.8% 1.771e+09 ± 2% -30.1% 5.339e+08 ± 2% hackbench.time.voluntary_context_switches
4051 ± 0% -39.9% 2434 ± 3% +57.8% 6393 ± 0% uptime.idle
4337715 ± 1% +7.3% 4654464 ± 2% -23.3% 3325346 ± 5% softirqs.RCU
2462880 ± 0% -35.6% 1585869 ± 5% +58.1% 3893988 ± 0% softirqs.SCHED
1766752 ± 1% +122.6% 3932589 ± 1% -25.6% 1313619 ± 1% vmstat.system.cs
249718 ± 2% +307.4% 1017398 ± 3% -40.4% 148723 ± 5% vmstat.system.in
1.76e+08 ± 3% +236.0% 5.914e+08 ± 2% -49.6% 88783232 ± 5% time.involuntary_context_switches
208664 ± 2% +26.0% 262929 ± 3% +15.7% 241377 ± 0% time.minor_page_faults
1272 ± 0% -24.5% 961.37 ± 2% -10.4% 1140 ± 0% time.user_time
7.64e+08 ± 1% +131.8% 1.771e+09 ± 2% -30.1% 5.339e+08 ± 2% time.voluntary_context_switches
177383 ± 0% +2.0% 180939 ± 0% -51.3% 86390 ± 1% meminfo.Active
102033 ± 0% -0.1% 101893 ± 1% -85.6% 14740 ± 0% meminfo.Active(file)
392558 ± 0% +0.0% 392612 ± 0% +22.6% 481411 ± 0% meminfo.Inactive
382911 ± 0% +0.0% 382923 ± 0% +23.2% 471792 ± 0% meminfo.Inactive(file)
143370 ± 0% -12.0% 126124 ± 1% -1.5% 141210 ± 0% meminfo.SUnreclaim
1136461 ± 3% +16.6% 1324662 ± 5% +15.9% 1316829 ± 1% numa-numastat.node0.local_node
1140216 ± 3% +16.2% 1324689 ± 5% +15.5% 1316830 ± 1% numa-numastat.node0.numa_hit
3755 ± 68% -99.3% 27.25 ± 94% -100.0% 1.25 ± 34% numa-numastat.node0.other_node
1098889 ± 4% +20.1% 1320211 ± 6% +16.4% 1278783 ± 1% numa-numastat.node1.local_node
1101996 ± 4% +20.5% 1327590 ± 6% +16.0% 1278783 ± 1% numa-numastat.node1.numa_hit
3106 ± 99% +137.5% 7379 ± 17% -100.0% 0.00 ± -1% numa-numastat.node1.other_node
7.18 ± 0% -50.2% 3.57 ± 43% +76.1% 12.64 ± 1% perf-profile.cycles-pp.call_cpuidle
8.09 ± 0% -44.7% 4.47 ± 38% +72.4% 13.95 ± 1% perf-profile.cycles-pp.cpu_startup_entry
7.17 ± 0% -50.3% 3.56 ± 43% +76.2% 12.63 ± 1% perf-profile.cycles-pp.cpuidle_enter
7.14 ± 0% -50.3% 3.55 ± 43% +76.1% 12.58 ± 1% perf-profile.cycles-pp.cpuidle_enter_state
7.11 ± 0% -50.6% 3.52 ± 43% +76.3% 12.54 ± 1% perf-profile.cycles-pp.intel_idle
8.00 ± 0% -44.5% 4.44 ± 38% +72.1% 13.77 ± 1% perf-profile.cycles-pp.start_secondary
92.32 ± 0% +5.4% 97.32 ± 0% -7.7% 85.26 ± 0% turbostat.%Busy
2763 ± 0% +5.4% 2912 ± 0% -7.7% 2551 ± 0% turbostat.Avg_MHz
7.48 ± 0% -66.5% 2.50 ± 7% +94.5% 14.54 ± 0% turbostat.CPU%c1
0.20 ± 2% -6.4% 0.18 ± 2% +2.6% 0.20 ± 3% turbostat.CPU%c6
180.03 ± 0% -1.3% 177.62 ± 0% -2.4% 175.63 ± 0% turbostat.CorWatt
209.86 ± 0% -0.8% 208.08 ± 0% -2.0% 205.64 ± 0% turbostat.PkgWatt
5.83 ± 0% +38.9% 8.10 ± 3% +12.7% 6.57 ± 1% turbostat.RAMWatt
1.658e+09 ± 0% -59.1% 6.784e+08 ± 7% +89.3% 3.138e+09 ± 0% cpuidle.C1-IVT.time
1.066e+08 ± 0% -40.3% 63661563 ± 6% +44.3% 1.539e+08 ± 0% cpuidle.C1-IVT.usage
26348635 ± 0% -86.8% 3471048 ± 15% +50.0% 39513523 ± 0% cpuidle.C1E-IVT.time
291620 ± 0% -85.1% 43352 ± 15% +28.8% 375730 ± 1% cpuidle.C1E-IVT.usage
54158643 ± 1% -88.5% 6254009 ± 14% +78.4% 96596486 ± 1% cpuidle.C3-IVT.time
482437 ± 1% -87.0% 62620 ± 16% +45.6% 702258 ± 1% cpuidle.C3-IVT.usage
5.028e+08 ± 0% -75.8% 1.219e+08 ± 8% +85.5% 9.327e+08 ± 1% cpuidle.C6-IVT.time
3805026 ± 0% -85.5% 552326 ± 16% +49.4% 5684182 ± 1% cpuidle.C6-IVT.usage
2766 ± 4% -51.4% 1344 ± 6% +10.0% 3042 ± 7% cpuidle.POLL.usage
49725 ± 4% +2.1% 50775 ± 3% -85.2% 7360 ± 0% numa-meminfo.node0.Active(file)
2228 ± 92% +137.1% 5285 ± 15% +118.7% 4874 ± 19% numa-meminfo.node0.AnonHugePages
197699 ± 2% +1.6% 200772 ± 0% +23.9% 245042 ± 0% numa-meminfo.node0.Inactive
192790 ± 1% -0.6% 191611 ± 0% +22.3% 235849 ± 0% numa-meminfo.node0.Inactive(file)
73589 ± 4% -12.5% 64393 ± 2% -1.3% 72664 ± 2% numa-meminfo.node0.SUnreclaim
27438 ± 83% +102.6% 55585 ± 6% +83.0% 50223 ± 0% numa-meminfo.node0.Shmem
101051 ± 3% -10.9% 90044 ± 2% -1.2% 99863 ± 2% numa-meminfo.node0.Slab
89204 ± 25% -25.3% 66594 ± 4% -77.6% 19954 ± 4% numa-meminfo.node1.Active
52306 ± 3% -2.3% 51117 ± 4% -85.9% 7380 ± 0% numa-meminfo.node1.Active(file)
194864 ± 2% -1.6% 191824 ± 1% +21.3% 236372 ± 0% numa-meminfo.node1.Inactive
4742 ± 86% -89.2% 511.75 ± 41% -90.9% 430.00 ± 60% numa-meminfo.node1.Inactive(anon)
190121 ± 1% +0.6% 191311 ± 1% +24.1% 235942 ± 0% numa-meminfo.node1.Inactive(file)
69844 ± 4% -11.8% 61579 ± 3% -1.9% 68521 ± 3% numa-meminfo.node1.SUnreclaim
12430 ± 4% +2.1% 12693 ± 3% -85.2% 1839 ± 0% numa-vmstat.node0.nr_active_file
48197 ± 1% -0.6% 47902 ± 0% +22.3% 58962 ± 0% numa-vmstat.node0.nr_inactive_file
6857 ± 83% +102.8% 13905 ± 6% +83.1% 12559 ± 0% numa-vmstat.node0.nr_shmem
18395 ± 4% -12.4% 16121 ± 2% -1.1% 18187 ± 2% numa-vmstat.node0.nr_slab_unreclaimable
675569 ± 3% +12.7% 761135 ± 4% +18.8% 802726 ± 4% numa-vmstat.node0.numa_local
71537 ± 5% -7.9% 65920 ± 2% -100.0% 0.25 ±173% numa-vmstat.node0.numa_other
13076 ± 3% -2.3% 12778 ± 4% -85.9% 1844 ± 0% numa-vmstat.node1.nr_active_file
1187 ± 86% -89.3% 127.50 ± 41% -91.0% 107.25 ± 60% numa-vmstat.node1.nr_inactive_anon
47530 ± 1% +0.6% 47827 ± 1% +24.1% 58985 ± 0% numa-vmstat.node1.nr_inactive_file
17456 ± 4% -11.7% 15405 ± 3% -1.9% 17127 ± 3% numa-vmstat.node1.nr_slab_unreclaimable
695848 ± 3% +14.9% 799683 ± 5% +4.7% 728368 ± 3% numa-vmstat.node1.numa_hit
677405 ± 4% +14.5% 775903 ± 6% +7.5% 728368 ± 3% numa-vmstat.node1.numa_local
18442 ± 19% +28.9% 23779 ± 5% -100.0% 0.00 ± -1% numa-vmstat.node1.numa_other
25508 ± 0% -0.1% 25473 ± 1% -85.6% 3684 ± 0% proc-vmstat.nr_active_file
95727 ± 0% +0.0% 95730 ± 0% +23.2% 117947 ± 0% proc-vmstat.nr_inactive_file
35841 ± 0% -12.0% 31543 ± 0% -1.5% 35298 ± 0% proc-vmstat.nr_slab_unreclaimable
154090 ± 2% +43.1% 220509 ± 3% +23.5% 190284 ± 0% proc-vmstat.numa_hint_faults
129240 ± 2% +47.4% 190543 ± 3% +15.1% 148733 ± 1% proc-vmstat.numa_hint_faults_local
2238386 ± 1% +18.4% 2649737 ± 2% +15.8% 2591197 ± 0% proc-vmstat.numa_hit
2232163 ± 1% +18.4% 2643105 ± 2% +16.1% 2591195 ± 0% proc-vmstat.numa_local
6223 ± 0% +6.6% 6632 ± 10% -100.0% 1.25 ± 34% proc-vmstat.numa_other
22315 ± 1% -21.0% 17625 ± 5% -0.4% 22234 ± 0% proc-vmstat.numa_pages_migrated
154533 ± 2% +45.6% 225071 ± 3% +25.7% 194235 ± 0% proc-vmstat.numa_pte_updates
14224 ± 0% +5.5% 15006 ± 3% -17.8% 11689 ± 0% proc-vmstat.pgactivate
382980 ± 2% +33.2% 510157 ± 4% +22.0% 467358 ± 0% proc-vmstat.pgalloc_dma32
7311738 ± 2% +37.2% 10029060 ± 2% +28.2% 9374740 ± 0% proc-vmstat.pgalloc_normal
7672040 ± 2% +37.1% 10519738 ± 2% +28.0% 9823026 ± 0% proc-vmstat.pgfree
22315 ± 1% -21.0% 17625 ± 5% -0.4% 22234 ± 0% proc-vmstat.pgmigrate_success
720.75 ± 3% -11.3% 639.50 ± 1% -29.2% 510.00 ± 0% slabinfo.RAW.active_objs
720.75 ± 3% -11.3% 639.50 ± 1% -29.2% 510.00 ± 0% slabinfo.RAW.num_objs
5487 ± 6% -12.6% 4797 ± 4% -100.0% 0.00 ± -1% slabinfo.UNIX.active_objs
164.50 ± 5% -12.3% 144.25 ± 4% -100.0% 0.00 ± -1% slabinfo.UNIX.active_slabs
5609 ± 5% -12.2% 4926 ± 4% -100.0% 0.00 ± -1% slabinfo.UNIX.num_objs
164.50 ± 5% -12.3% 144.25 ± 4% -100.0% 0.00 ± -1% slabinfo.UNIX.num_slabs
4362 ± 4% +14.6% 4998 ± 2% -3.2% 4223 ± 4% slabinfo.cred_jar.active_objs
4362 ± 4% +14.6% 4998 ± 2% -3.2% 4223 ± 4% slabinfo.cred_jar.num_objs
2904 ± 4% -2.7% 2825 ± 1% +56.5% 4545 ± 2% slabinfo.kmalloc-1024.active_objs
2935 ± 2% -0.5% 2920 ± 1% +57.8% 4633 ± 2% slabinfo.kmalloc-1024.num_objs
42525 ± 0% -41.6% 24824 ± 3% +7.3% 45621 ± 0% slabinfo.kmalloc-256.active_objs
845.50 ± 0% -42.9% 482.50 ± 3% +3.0% 870.50 ± 0% slabinfo.kmalloc-256.active_slabs
54124 ± 0% -42.9% 30920 ± 3% +3.0% 55755 ± 0% slabinfo.kmalloc-256.num_objs
845.50 ± 0% -42.9% 482.50 ± 3% +3.0% 870.50 ± 0% slabinfo.kmalloc-256.num_slabs
47204 ± 0% -37.9% 29335 ± 2% +6.6% 50334 ± 0% slabinfo.kmalloc-512.active_objs
915.25 ± 0% -39.8% 551.00 ± 3% +2.8% 940.50 ± 0% slabinfo.kmalloc-512.active_slabs
58599 ± 0% -39.8% 35300 ± 3% +2.8% 60224 ± 0% slabinfo.kmalloc-512.num_objs
915.25 ± 0% -39.8% 551.00 ± 3% +2.8% 940.50 ± 0% slabinfo.kmalloc-512.num_slabs
12443 ± 2% -20.1% 9944 ± 3% -6.5% 11639 ± 1% slabinfo.pid.active_objs
12443 ± 2% -20.1% 9944 ± 3% -6.5% 11639 ± 1% slabinfo.pid.num_objs
440.00 ± 5% -32.8% 295.75 ± 4% -11.7% 388.50 ± 7% slabinfo.taskstats.active_objs
440.00 ± 5% -32.8% 295.75 ± 4% -11.7% 388.50 ± 7% slabinfo.taskstats.num_objs
188235 ± 74% +62.9% 306699 ± 27% -98.6% 2627 ± 40% sched_debug.cfs_rq:/.MIN_vruntime.avg
7146629 ± 80% +27.7% 9122933 ± 36% -98.6% 98261 ± 36% sched_debug.cfs_rq:/.MIN_vruntime.max
1117852 ± 77% +44.7% 1617052 ± 31% -98.6% 15548 ± 37% sched_debug.cfs_rq:/.MIN_vruntime.stddev
61.52 ±116% -70.6% 18.11 ± 6% +1.2e+06% 718736 ± 1% sched_debug.cfs_rq:/.load.avg
2144 ±161% -96.3% 79.41 ± 48% +49309.2% 1059411 ± 3% sched_debug.cfs_rq:/.load.max
312.45 ±157% -94.8% 16.29 ± 33% +1.1e+05% 333106 ± 5% sched_debug.cfs_rq:/.load.stddev
20.46 ± 4% +9.0% 22.31 ± 6% +3004.0% 635.15 ± 1% sched_debug.cfs_rq:/.load_avg.avg
81.57 ± 32% +14.2% 93.18 ± 26% +1035.5% 926.18 ± 3% sched_debug.cfs_rq:/.load_avg.max
8.14 ± 5% -2.8% 7.91 ± 3% +2585.8% 218.52 ± 13% sched_debug.cfs_rq:/.load_avg.min
13.90 ± 29% +16.9% 16.25 ± 22% +1089.3% 165.34 ± 5% sched_debug.cfs_rq:/.load_avg.stddev
188235 ± 74% +62.9% 306699 ± 27% -98.6% 2627 ± 40% sched_debug.cfs_rq:/.max_vruntime.avg
7146629 ± 80% +27.7% 9122933 ± 36% -98.6% 98261 ± 36% sched_debug.cfs_rq:/.max_vruntime.max
1117852 ± 77% +44.7% 1617052 ± 31% -98.6% 15548 ± 37% sched_debug.cfs_rq:/.max_vruntime.stddev
29491781 ± 0% -4.8% 28074842 ± 1% -99.0% 295426 ± 0% sched_debug.cfs_rq:/.min_vruntime.avg
31241540 ± 0% -5.8% 29418054 ± 0% -99.0% 320734 ± 0% sched_debug.cfs_rq:/.min_vruntime.max
27849652 ± 0% -3.7% 26821072 ± 2% -99.0% 275550 ± 0% sched_debug.cfs_rq:/.min_vruntime.min
861989 ± 3% -20.2% 687639 ± 22% -98.3% 14586 ± 2% sched_debug.cfs_rq:/.min_vruntime.stddev
0.27 ± 5% -56.3% 0.12 ± 30% +27.5% 0.34 ± 6% sched_debug.cfs_rq:/.nr_running.stddev
16.51 ± 1% +9.5% 18.08 ± 3% +3343.1% 568.61 ± 2% sched_debug.cfs_rq:/.runnable_load_avg.avg
34.80 ± 13% +15.0% 40.02 ± 19% +2514.0% 909.57 ± 0% sched_debug.cfs_rq:/.runnable_load_avg.max
0.05 ±100% +7950.0% 3.66 ± 48% +3250.0% 1.52 ± 89% sched_debug.cfs_rq:/.runnable_load_avg.min
7.18 ± 9% -0.1% 7.18 ± 13% +3571.2% 263.68 ± 4% sched_debug.cfs_rq:/.runnable_load_avg.stddev
-740916 ±-28% -158.5% 433310 ±120% -96.8% -23579 ± -5% sched_debug.cfs_rq:/.spread0.avg
1009940 ± 19% +75.8% 1775442 ± 30% -99.8% 1736 ±164% sched_debug.cfs_rq:/.spread0.max
-2384171 ± -7% -65.7% -818684 ±-76% -98.2% -43456 ± -4% sched_debug.cfs_rq:/.spread0.min
862765 ± 3% -20.4% 686825 ± 22% -98.3% 14591 ± 2% sched_debug.cfs_rq:/.spread0.stddev
749.14 ± 1% +13.0% 846.34 ± 1% -41.1% 441.05 ± 5% sched_debug.cfs_rq:/.util_avg.min
51.66 ± 4% -36.3% 32.92 ± 5% +150.6% 129.46 ± 6% sched_debug.cfs_rq:/.util_avg.stddev
161202 ± 7% -41.7% 93997 ± 4% +147.7% 399342 ± 1% sched_debug.cpu.avg_idle.avg
595158 ± 6% -51.2% 290491 ± 22% +37.8% 820120 ± 0% sched_debug.cpu.avg_idle.max
7658 ± 51% +9.2% 8366 ± 26% +114.4% 16423 ± 31% sched_debug.cpu.avg_idle.min
132760 ± 8% -58.8% 54718 ± 19% +97.8% 262608 ± 0% sched_debug.cpu.avg_idle.stddev
11.40 ± 11% +111.0% 24.05 ± 16% -58.1% 4.78 ± 3% sched_debug.cpu.clock.stddev
11.40 ± 11% +111.0% 24.05 ± 16% -58.1% 4.78 ± 3% sched_debug.cpu.clock_task.stddev
16.59 ± 1% +7.7% 17.86 ± 2% +3099.8% 530.73 ± 2% sched_debug.cpu.cpu_load[0].avg
32.34 ± 2% +23.9% 40.07 ± 19% +2715.0% 910.41 ± 0% sched_debug.cpu.cpu_load[0].max
0.34 ±103% +520.0% 2.11 ± 67% +140.0% 0.82 ±110% sched_debug.cpu.cpu_load[0].min
6.87 ± 3% +8.0% 7.42 ± 13% +4228.9% 297.50 ± 3% sched_debug.cpu.cpu_load[0].stddev
16.56 ± 0% +8.1% 17.91 ± 2% +3703.9% 630.04 ± 1% sched_debug.cpu.cpu_load[1].avg
32.18 ± 2% +22.7% 39.50 ± 17% +2728.5% 910.25 ± 0% sched_debug.cpu.cpu_load[1].max
3.32 ± 8% +84.9% 6.14 ± 12% +5364.4% 181.32 ± 9% sched_debug.cpu.cpu_load[1].min
6.14 ± 5% +12.5% 6.91 ± 13% +2708.6% 172.56 ± 5% sched_debug.cpu.cpu_load[1].stddev
16.75 ± 1% +7.6% 18.02 ± 2% +3646.9% 627.69 ± 1% sched_debug.cpu.cpu_load[2].avg
33.25 ± 7% +16.5% 38.75 ± 14% +2634.1% 909.09 ± 0% sched_debug.cpu.cpu_load[2].max
5.39 ± 7% +36.3% 7.34 ± 4% +3547.3% 196.45 ± 11% sched_debug.cpu.cpu_load[2].min
5.95 ± 9% +11.8% 6.65 ± 11% +2752.1% 169.73 ± 5% sched_debug.cpu.cpu_load[2].stddev
17.17 ± 1% +6.1% 18.22 ± 2% +3552.1% 626.96 ± 1% sched_debug.cpu.cpu_load[3].avg
33.20 ± 7% +14.6% 38.05 ± 9% +2631.3% 906.93 ± 0% sched_debug.cpu.cpu_load[3].max
6.93 ± 7% +10.5% 7.66 ± 1% +2766.9% 198.73 ± 11% sched_debug.cpu.cpu_load[3].min
5.70 ± 9% +13.9% 6.49 ± 8% +2825.6% 166.73 ± 5% sched_debug.cpu.cpu_load[3].stddev
17.49 ± 0% +4.9% 18.36 ± 2% +3482.1% 626.64 ± 1% sched_debug.cpu.cpu_load[4].avg
33.18 ± 3% +14.0% 37.82 ± 5% +2615.8% 901.16 ± 0% sched_debug.cpu.cpu_load[4].max
7.66 ± 8% +0.9% 7.73 ± 1% +2568.8% 204.41 ± 11% sched_debug.cpu.cpu_load[4].min
5.56 ± 6% +16.2% 6.45 ± 6% +2814.9% 161.96 ± 6% sched_debug.cpu.cpu_load[4].stddev
16741 ± 0% -15.4% 14166 ± 2% -13.0% 14564 ± 2% sched_debug.cpu.curr->pid.avg
19196 ± 0% -18.3% 15690 ± 1% -4.9% 18255 ± 0% sched_debug.cpu.curr->pid.max
5174 ± 5% -55.4% 2305 ± 14% +19.3% 6173 ± 6% sched_debug.cpu.curr->pid.stddev
18.60 ± 5% -2.7% 18.10 ± 6% +3.9e+06% 717646 ± 2% sched_debug.cpu.load.avg
81.23 ± 48% -2.4% 79.30 ± 47% +1.3e+06% 1059340 ± 3% sched_debug.cpu.load.max
18.01 ± 28% -9.4% 16.32 ± 33% +1.9e+06% 333436 ± 5% sched_debug.cpu.load.stddev
0.00 ± 2% +29.8% 0.00 ± 33% +39.0% 0.00 ± 15% sched_debug.cpu.next_balance.stddev
1410 ± 1% -14.2% 1210 ± 6% +34.5% 1896 ± 1% sched_debug.cpu.nr_load_updates.stddev
9.95 ± 3% -14.5% 8.51 ± 5% -1.2% 9.83 ± 2% sched_debug.cpu.nr_running.avg
29.07 ± 2% -15.0% 24.70 ± 4% +37.5% 39.98 ± 1% sched_debug.cpu.nr_running.max
0.05 ±100% +850.0% 0.43 ± 37% -100.0% 0.00 ± -1% sched_debug.cpu.nr_running.min
7.64 ± 3% -23.0% 5.88 ± 2% +48.6% 11.36 ± 2% sched_debug.cpu.nr_running.stddev
10979930 ± 1% +123.3% 24518490 ± 2% -26.3% 8091669 ± 1% sched_debug.cpu.nr_switches.avg
12350130 ± 1% +117.5% 26856375 ± 2% -17.0% 10249081 ± 2% sched_debug.cpu.nr_switches.max
9594835 ± 2% +132.6% 22314436 ± 2% -31.0% 6620975 ± 2% sched_debug.cpu.nr_switches.min
769296 ± 1% +56.8% 1206190 ± 3% +54.6% 1189172 ± 1% sched_debug.cpu.nr_switches.stddev
8.30 ± 18% +32.9% 11.02 ± 15% +113.7% 17.73 ± 26% sched_debug.cpu.nr_uninterruptible.max
4.87 ± 15% +14.3% 5.57 ± 6% +97.2% 9.61 ± 29% sched_debug.cpu.nr_uninterruptible.stddev
[toc] | [next] | [standalone]
| From | Peter Zijlstra <peterz@infradead.org> |
|---|---|
| Date | 2016-05-31 14:50 +0200 |
| Subject | Re: [LKP] [lkp] [sched/fair] 53d3bc773e: hackbench.throughput -32.9% regression |
| Message-ID | <rERho-2Kk-5@gated-at.bofh.it> |
| In reply to | #1409991 |
On Tue, May 31, 2016 at 04:34:36PM +0800, Huang, Ying wrote:
> Hi, Ingo,
>
> Part of the regression has been recovered in v4.7-rc1 from -32.9% to
> -9.8%. But there is still some regression. Is it possible for fully
> restore it?
after much searching on how you guys run hackbench... I figured
something like:
perf bench sched messaging -g 20 --thread -l 60000
on my IVB-EP (2*10*2) is similar to your IVT thing.
And running something like:
for i in /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor ; do echo performance > $i ; done
perf stat --null --repeat 10 -- perf bench sched messaging -g 20 --thread -l 60000 | grep "seconds time elapsed"
gets me:
v4.6:
36.786914089 seconds time elapsed ( +- 0.49% )
37.054017355 seconds time elapsed ( +- 1.05% )
origin/master (v4.7-rc1-ish):
34.757435264 seconds time elapsed ( +- 3.34% )
35.396252515 seconds time elapsed ( +- 3.38% )
Which doesn't show a regression between v4.6 and HEAD; in fact it shows
an improvement.
[toc] | [prev] | [next] | [standalone]
| From | "Huang\, Ying" <ying.huang@intel.com> |
|---|---|
| Date | 2016-06-01 07:10 +0200 |
| Message-ID | <rF6zL-4aY-1@gated-at.bofh.it> |
| In reply to | #1410267 |
Hi, Peter,
Peter Zijlstra <peterz@infradead.org> writes:
> On Tue, May 31, 2016 at 04:34:36PM +0800, Huang, Ying wrote:
>> Hi, Ingo,
>>
>> Part of the regression has been recovered in v4.7-rc1 from -32.9% to
>> -9.8%. But there is still some regression. Is it possible for fully
>> restore it?
>
> after much searching on how you guys run hackbench... I figured
> something like:
>
> perf bench sched messaging -g 20 --thread -l 60000
There is a reproduce file attached in the original report email, its
contents is something like below:
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu1/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu10/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu11/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu12/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu13/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu14/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu15/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu16/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu17/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu18/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu19/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu2/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu20/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu21/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu22/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu23/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu24/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu25/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu26/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu27/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu28/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu29/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu3/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu30/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu31/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu32/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu33/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu34/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu35/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu36/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu37/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu38/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu39/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu4/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu40/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu41/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu42/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu43/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu44/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu45/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu46/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu47/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu5/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu6/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu7/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu8/cpufreq/scaling_governor
2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu9/cpufreq/scaling_governor
2016-05-15 08:57:03 /usr/bin/hackbench -g 24 --threads -l 60000
2016-05-15 08:57:50 /usr/bin/hackbench -g 24 --threads -l 60000
2016-05-15 08:58:33 /usr/bin/hackbench -g 24 --threads -l 60000
2016-05-15 08:59:15 /usr/bin/hackbench -g 24 --threads -l 60000
2016-05-15 08:59:58 /usr/bin/hackbench -g 24 --threads -l 60000
2016-05-15 09:00:43 /usr/bin/hackbench -g 24 --threads -l 60000
2016-05-15 09:01:22 /usr/bin/hackbench -g 24 --threads -l 60000
2016-05-15 09:01:57 /usr/bin/hackbench -g 24 --threads -l 60000
2016-05-15 09:02:39 /usr/bin/hackbench -g 24 --threads -l 60000
2016-05-15 09:03:22 /usr/bin/hackbench -g 24 --threads -l 60000
2016-05-15 09:04:10 /usr/bin/hackbench -g 24 --threads -l 60000
2016-05-15 09:04:53 /usr/bin/hackbench -g 24 --threads -l 60000
2016-05-15 09:05:39 /usr/bin/hackbench -g 24 --threads -l 60000
2016-05-15 09:06:24 /usr/bin/hackbench -g 24 --threads -l 60000
Hope that will help you for reproduce.
> on my IVB-EP (2*10*2) is similar to your IVT thing.
>
> And running something like:
>
> for i in /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor ; do echo performance > $i ; done
> perf stat --null --repeat 10 -- perf bench sched messaging -g 20 --thread -l 60000 | grep "seconds time elapsed"
>
> gets me:
>
> v4.6:
>
> 36.786914089 seconds time elapsed ( +- 0.49% )
> 37.054017355 seconds time elapsed ( +- 1.05% )
>
>
> origin/master (v4.7-rc1-ish):
>
> 34.757435264 seconds time elapsed ( +- 3.34% )
> 35.396252515 seconds time elapsed ( +- 3.38% )
>
>
> Which doesn't show a regression between v4.6 and HEAD; in fact it shows
> an improvement.
Yes. For hackbench test, linus/master (v4.7-rc1+) is better than v4.6,
but it is worse than v4.6-rc7. Details is as below.
=========================================================================================
tbox_group/testcase/rootfs/kconfig/compiler/cpufreq_governor/nr_threads/mode/ipc:
ivb42/hackbench/debian-x86_64-2015-02-07.cgz/x86_64-rhel/gcc-4.9/performance/50%/threads/socket
commit:
v4.6-rc7
v4.6
367d3fd50566a313946fa9c5b2116a81bf3807e4
v4.6-rc7 v4.6 367d3fd50566a313946fa9c5b2
---------------- -------------------------- --------------------------
%stddev %change %stddev %change %stddev
\ | \ | \
198307 ± 0% -33.4% 132165 ± 3% -11.8% 174857 ± 0% hackbench.throughput
625.91 ± 0% -2.0% 613.12 ± 1% -2.1% 612.85 ± 0% hackbench.time.elapsed_time
625.91 ± 0% -2.0% 613.12 ± 1% -2.1% 612.85 ± 0% hackbench.time.elapsed_time.max
1.611e+08 ± 0% +254.7% 5.712e+08 ± 4% -25.3% 1.203e+08 ± 5% hackbench.time.involuntary_context_switches
212287 ± 2% +22.3% 259622 ± 4% +33.0% 282261 ± 1% hackbench.time.minor_page_faults
4391 ± 0% +5.7% 4643 ± 0% -6.9% 4090 ± 0% hackbench.time.percent_of_cpu_this_job_got
26154 ± 0% +5.2% 27509 ± 1% -8.5% 23935 ± 0% hackbench.time.system_time
1336 ± 0% -28.1% 961.07 ± 2% -14.8% 1138 ± 0% hackbench.time.user_time
7.442e+08 ± 0% +129.6% 1.709e+09 ± 4% -17.5% 6.139e+08 ± 2% hackbench.time.voluntary_context_switches
4157 ± 1% -39.0% 2536 ± 15% +44.6% 6011 ± 2% uptime.idle
1656569 ± 0% +131.8% 3840033 ± 3% -10.2% 1486840 ± 2% vmstat.system.cs
225682 ± 0% +335.2% 982245 ± 5% -4.2% 216300 ± 7% vmstat.system.in
4416560 ± 3% +7.3% 4737257 ± 2% -18.1% 3617836 ± 1% softirqs.RCU
2591680 ± 0% -37.9% 1608431 ± 7% +47.9% 3833673 ± 0% softirqs.SCHED
13948275 ± 0% +3.3% 14406201 ± 1% -8.9% 12703887 ± 0% softirqs.TIMER
1.611e+08 ± 0% +254.7% 5.712e+08 ± 4% -25.3% 1.203e+08 ± 5% time.involuntary_context_switches
212287 ± 2% +22.3% 259622 ± 4% +33.0% 282261 ± 1% time.minor_page_faults
1336 ± 0% -28.1% 961.07 ± 2% -14.8% 1138 ± 0% time.user_time
7.442e+08 ± 0% +129.6% 1.709e+09 ± 4% -17.5% 6.139e+08 ± 2% time.voluntary_context_switches
176970 ± 1% +2.4% 181276 ± 0% -51.5% 85865 ± 0% meminfo.Active
101149 ± 2% +0.4% 101589 ± 1% -85.4% 14807 ± 0% meminfo.Active(file)
390916 ± 0% +1.1% 395022 ± 0% +23.2% 481664 ± 0% meminfo.Inactive
381267 ± 0% +1.1% 385296 ± 0% +23.8% 472035 ± 0% meminfo.Inactive(file)
143716 ± 0% -12.4% 125923 ± 1% -2.4% 140230 ± 0% meminfo.SUnreclaim
194906 ± 0% -8.9% 177650 ± 1% -1.8% 191478 ± 0% meminfo.Slab
1162151 ± 6% +11.4% 1294775 ± 2% +17.5% 1365360 ± 1% numa-numastat.node0.local_node
1163400 ± 6% +11.5% 1297646 ± 2% +17.4% 1365361 ± 1% numa-numastat.node0.numa_hit
1249 ±197% +129.8% 2871 ± 95% -99.9% 0.67 ± 70% numa-numastat.node0.other_node
1084104 ± 6% +15.1% 1247352 ± 4% +22.0% 1323149 ± 1% numa-numastat.node1.local_node
1089973 ± 6% +14.9% 1252683 ± 4% +21.4% 1323149 ± 1% numa-numastat.node1.numa_hit
5868 ± 40% -9.2% 5330 ± 70% -100.0% 0.33 ±141% numa-numastat.node1.other_node
92.11 ± 0% +5.5% 97.16 ± 0% -6.3% 86.33 ± 0% turbostat.%Busy
2756 ± 0% +5.5% 2907 ± 0% -6.3% 2584 ± 0% turbostat.Avg_MHz
7.70 ± 0% -65.6% 2.64 ± 12% +74.9% 13.46 ± 2% turbostat.CPU%c1
180.27 ± 0% -1.6% 177.34 ± 0% -2.5% 175.80 ± 0% turbostat.CorWatt
210.07 ± 0% -1.1% 207.71 ± 0% -1.9% 206.01 ± 0% turbostat.PkgWatt
5.81 ± 0% +35.8% 7.88 ± 3% +24.2% 7.21 ± 2% turbostat.RAMWatt
102504 ± 20% -5.6% 96726 ± 25% -65.7% 35129 ± 52% numa-meminfo.node0.Active
50026 ± 2% +2.3% 51197 ± 4% -85.2% 7408 ± 0% numa-meminfo.node0.Active(file)
198553 ± 2% +0.4% 199265 ± 3% +22.0% 242211 ± 1% numa-meminfo.node0.Inactive
191148 ± 1% +1.7% 194350 ± 3% +23.5% 235978 ± 0% numa-meminfo.node0.Inactive(file)
74572 ± 8% -11.8% 65807 ± 5% -4.4% 71257 ± 3% numa-meminfo.node0.SUnreclaim
51121 ± 5% -1.4% 50391 ± 4% -85.5% 7398 ± 0% numa-meminfo.node1.Active(file)
192353 ± 1% +1.8% 195730 ± 2% +24.5% 239430 ± 1% numa-meminfo.node1.Inactive
190119 ± 0% +0.4% 190946 ± 1% +24.2% 236055 ± 0% numa-meminfo.node1.Inactive(file)
472112 ± 5% +3.0% 486190 ± 5% +8.2% 510902 ± 4% numa-meminfo.node1.MemUsed
12506 ± 2% +2.3% 12799 ± 4% -85.2% 1852 ± 0% numa-vmstat.node0.nr_active_file
47786 ± 1% +1.7% 48587 ± 3% +23.5% 58994 ± 0% numa-vmstat.node0.nr_inactive_file
18626 ± 8% -11.7% 16446 ± 5% -4.4% 17801 ± 3% numa-vmstat.node0.nr_slab_unreclaimable
66037 ± 3% +3.1% 68095 ± 4% -100.0% 0.00 ± 0% numa-vmstat.node0.numa_other
12780 ± 5% -1.4% 12597 ± 4% -85.5% 1849 ± 0% numa-vmstat.node1.nr_active_file
47529 ± 0% +0.4% 47735 ± 1% +24.2% 59013 ± 0% numa-vmstat.node1.nr_inactive_file
698206 ± 5% +11.3% 777438 ± 4% +17.6% 820805 ± 2% numa-vmstat.node1.numa_hit
674672 ± 6% +12.0% 755944 ± 4% +21.7% 820805 ± 2% numa-vmstat.node1.numa_local
23532 ± 10% -8.7% 21493 ± 15% -100.0% 0.00 ± 0% numa-vmstat.node1.numa_other
1.766e+09 ± 0% -60.1% 7.057e+08 ± 11% +70.1% 3.004e+09 ± 1% cpuidle.C1-IVT.time
1.125e+08 ± 0% -41.9% 65415380 ± 10% +38.6% 1.559e+08 ± 0% cpuidle.C1-IVT.usage
28400387 ± 1% -86.0% 3980259 ± 24% +21.9% 34611888 ± 3% cpuidle.C1E-IVT.time
308989 ± 0% -84.5% 47825 ± 23% +10.1% 340115 ± 3% cpuidle.C1E-IVT.usage
58891432 ± 0% -88.2% 6936400 ± 20% +36.2% 80209704 ± 4% cpuidle.C3-IVT.time
521047 ± 0% -86.5% 70085 ± 22% +16.6% 607626 ± 3% cpuidle.C3-IVT.usage
5.375e+08 ± 0% -75.8% 1.298e+08 ± 11% +55.6% 8.366e+08 ± 2% cpuidle.C6-IVT.time
4062211 ± 0% -85.1% 603908 ± 22% +28.0% 5200129 ± 2% cpuidle.C6-IVT.usage
15694 ± 6% +386.2% 76300 ±145% +774.3% 137212 ± 62% cpuidle.POLL.time
2751 ± 3% -52.5% 1308 ± 18% +15.4% 3176 ± 2% cpuidle.POLL.usage
25287 ± 2% +0.4% 25397 ± 1% -85.4% 3701 ± 0% proc-vmstat.nr_active_file
95316 ± 0% +1.1% 96323 ± 0% +23.8% 118008 ± 0% proc-vmstat.nr_inactive_file
35930 ± 0% -12.3% 31511 ± 1% -2.5% 35048 ± 0% proc-vmstat.nr_slab_unreclaimable
154964 ± 3% +40.6% 217915 ± 5% +48.6% 230354 ± 2% proc-vmstat.numa_hint_faults
128683 ± 4% +46.4% 188443 ± 5% +45.5% 187179 ± 2% proc-vmstat.numa_hint_faults_local
2247802 ± 0% +13.2% 2544572 ± 2% +19.5% 2685990 ± 0% proc-vmstat.numa_hit
2241597 ± 0% +13.2% 2537511 ± 2% +19.8% 2685989 ± 0% proc-vmstat.numa_local
6205 ± 0% +13.8% 7060 ± 18% -100.0% 1.00 ± 0% proc-vmstat.numa_other
23151 ± 1% -25.8% 17189 ± 4% -1.7% 22762 ± 0% proc-vmstat.numa_pages_migrated
155763 ± 3% +43.4% 223408 ± 5% +49.7% 233247 ± 2% proc-vmstat.numa_pte_updates
14010 ± 1% +16.3% 16287 ± 7% -17.1% 11610 ± 0% proc-vmstat.pgactivate
373910 ± 2% +28.4% 479928 ± 4% +30.1% 486506 ± 1% proc-vmstat.pgalloc_dma32
7157922 ± 1% +30.9% 9370533 ± 2% +38.0% 9878095 ± 0% proc-vmstat.pgalloc_normal
7509133 ± 1% +30.9% 9827974 ± 2% +37.8% 10345598 ± 0% proc-vmstat.pgfree
23151 ± 1% -25.8% 17189 ± 4% -1.7% 22762 ± 0% proc-vmstat.pgmigrate_success
737.40 ± 4% -10.3% 661.25 ± 3% -30.8% 510.00 ± 0% slabinfo.RAW.active_objs
737.40 ± 4% -10.3% 661.25 ± 3% -30.8% 510.00 ± 0% slabinfo.RAW.num_objs
5762 ± 6% -19.2% 4653 ± 3% -100.0% 0.00 ± -1% slabinfo.UNIX.active_objs
172.60 ± 6% -18.9% 140.00 ± 3% -100.0% 0.00 ± -1% slabinfo.UNIX.active_slabs
5892 ± 6% -19.0% 4775 ± 3% -100.0% 0.00 ± -1% slabinfo.UNIX.num_objs
172.60 ± 6% -18.9% 140.00 ± 3% -100.0% 0.00 ± -1% slabinfo.UNIX.num_slabs
37256 ± 3% -8.7% 34010 ± 3% +1.6% 37863 ± 0% slabinfo.anon_vma_chain.active_objs
37401 ± 3% -8.8% 34094 ± 3% +1.5% 37948 ± 0% slabinfo.anon_vma_chain.num_objs
4509 ± 1% +13.8% 5130 ± 9% +8.3% 4885 ± 15% slabinfo.cred_jar.active_objs
4509 ± 1% +13.8% 5130 ± 9% +8.3% 4885 ± 15% slabinfo.cred_jar.num_objs
2783 ± 2% +3.4% 2877 ± 4% +54.3% 4295 ± 0% slabinfo.kmalloc-1024.active_objs
2858 ± 1% +2.6% 2932 ± 3% +53.4% 4385 ± 0% slabinfo.kmalloc-1024.num_objs
25441 ± 1% -10.1% 22884 ± 1% -3.8% 24477 ± 2% slabinfo.kmalloc-16.active_objs
25441 ± 1% -10.1% 22884 ± 1% -3.8% 24477 ± 2% slabinfo.kmalloc-16.num_objs
43013 ± 0% -41.4% 25205 ± 5% +3.1% 44366 ± 1% slabinfo.kmalloc-256.active_objs
854.60 ± 0% -42.0% 495.25 ± 5% -1.0% 846.00 ± 0% slabinfo.kmalloc-256.active_slabs
54719 ± 0% -42.0% 31735 ± 5% -1.0% 54189 ± 0% slabinfo.kmalloc-256.num_objs
854.60 ± 0% -42.0% 495.25 ± 5% -1.0% 846.00 ± 0% slabinfo.kmalloc-256.num_slabs
47683 ± 0% -37.7% 29715 ± 4% +2.9% 49067 ± 0% slabinfo.kmalloc-512.active_objs
924.00 ± 0% -39.0% 563.75 ± 4% -0.9% 916.00 ± 0% slabinfo.kmalloc-512.active_slabs
59169 ± 0% -39.0% 36109 ± 4% -0.8% 58667 ± 0% slabinfo.kmalloc-512.num_objs
924.00 ± 0% -39.0% 563.75 ± 4% -0.9% 916.00 ± 0% slabinfo.kmalloc-512.num_slabs
8287 ± 2% +2.8% 8521 ± 4% +12.6% 9335 ± 2% slabinfo.kmalloc-96.active_objs
8351 ± 3% +2.6% 8570 ± 4% +12.7% 9409 ± 2% slabinfo.kmalloc-96.num_objs
12776 ± 1% -22.2% 9944 ± 2% -6.8% 11906 ± 1% slabinfo.pid.active_objs
12776 ± 1% -22.2% 9944 ± 2% -6.8% 11906 ± 1% slabinfo.pid.num_objs
5708 ± 2% -10.0% 5139 ± 3% -6.2% 5355 ± 0% slabinfo.sock_inode_cache.active_objs
5902 ± 2% -9.8% 5326 ± 3% -5.9% 5552 ± 0% slabinfo.sock_inode_cache.num_objs
447.40 ± 6% -35.7% 287.50 ± 6% -7.7% 413.00 ± 4% slabinfo.taskstats.active_objs
447.40 ± 6% -35.7% 287.50 ± 6% -7.7% 413.00 ± 4% slabinfo.taskstats.num_objs
304731 ± 27% -45.5% 166107 ± 76% -98.3% 5031 ± 23% sched_debug.cfs_rq:/.MIN_vruntime.avg
12211047 ± 35% -38.5% 7509311 ± 78% -99.0% 118856 ± 40% sched_debug.cfs_rq:/.MIN_vruntime.max
1877477 ± 30% -41.5% 1098508 ± 77% -98.8% 21976 ± 14% sched_debug.cfs_rq:/.MIN_vruntime.stddev
18.91 ± 7% -3.9% 18.16 ± 8% +3.8e+06% 715502 ± 2% sched_debug.cfs_rq:/.load.avg
95.71 ± 45% -7.8% 88.20 ± 74% +1.1e+06% 1067373 ± 4% sched_debug.cfs_rq:/.load.max
19.94 ± 31% -9.6% 18.02 ± 52% +1.7e+06% 335607 ± 2% sched_debug.cfs_rq:/.load.stddev
21.16 ± 9% +12.3% 23.76 ± 9% +2890.4% 632.65 ± 3% sched_debug.cfs_rq:/.load_avg.avg
125.40 ± 49% +4.4% 130.90 ± 13% +643.9% 932.88 ± 5% sched_debug.cfs_rq:/.load_avg.max
8.29 ± 2% -3.5% 8.00 ± 6% +2852.1% 244.76 ± 6% sched_debug.cfs_rq:/.load_avg.min
20.18 ± 45% +13.1% 22.83 ± 18% +720.8% 165.65 ± 3% sched_debug.cfs_rq:/.load_avg.stddev
304731 ± 27% -45.5% 166107 ± 76% -98.3% 5031 ± 23% sched_debug.cfs_rq:/.max_vruntime.avg
12211047 ± 35% -38.5% 7509311 ± 78% -99.0% 118856 ± 40% sched_debug.cfs_rq:/.max_vruntime.max
1877477 ± 30% -41.5% 1098508 ± 77% -98.8% 21976 ± 14% sched_debug.cfs_rq:/.max_vruntime.stddev
29445770 ± 0% -4.3% 28190370 ± 2% -99.0% 299502 ± 0% sched_debug.cfs_rq:/.min_vruntime.avg
31331918 ± 0% -6.2% 29380072 ± 2% -99.0% 322082 ± 0% sched_debug.cfs_rq:/.min_vruntime.max
27785446 ± 0% -2.5% 27098935 ± 2% -99.0% 282267 ± 0% sched_debug.cfs_rq:/.min_vruntime.min
916182 ± 13% -35.6% 590123 ± 13% -98.6% 12421 ± 4% sched_debug.cfs_rq:/.min_vruntime.stddev
0.26 ± 6% -34.5% 0.17 ± 14% +34.0% 0.34 ± 3% sched_debug.cfs_rq:/.nr_running.stddev
16.42 ± 3% +6.9% 17.56 ± 3% +3319.3% 561.57 ± 4% sched_debug.cfs_rq:/.runnable_load_avg.avg
38.22 ± 28% +10.9% 42.38 ± 49% +2280.8% 909.91 ± 1% sched_debug.cfs_rq:/.runnable_load_avg.max
0.05 ±133% +4879.2% 2.72 ± 46% +4177.8% 2.33 ± 32% sched_debug.cfs_rq:/.runnable_load_avg.min
7.59 ± 17% +4.0% 7.90 ± 36% +3375.3% 263.95 ± 1% sched_debug.cfs_rq:/.runnable_load_avg.stddev
-897515 ±-52% -132.1% 288533 ±159% -97.6% -21836 ± -6% sched_debug.cfs_rq:/.spread0.avg
989517 ± 31% +49.2% 1476487 ± 23% -99.9% 748.12 ±129% sched_debug.cfs_rq:/.spread0.max
-2558887 ±-23% -68.7% -801084 ±-66% -98.5% -39072 ± -7% sched_debug.cfs_rq:/.spread0.min
916967 ± 13% -35.7% 589208 ± 13% -98.6% 12424 ± 4% sched_debug.cfs_rq:/.spread0.stddev
744.20 ± 0% +10.9% 825.23 ± 3% -38.6% 457.27 ± 3% sched_debug.cfs_rq:/.util_avg.min
58.07 ± 9% -28.4% 41.55 ± 19% +119.1% 127.19 ± 2% sched_debug.cfs_rq:/.util_avg.stddev
157158 ± 3% -35.8% 100942 ± 9% +135.5% 370117 ± 7% sched_debug.cpu.avg_idle.avg
600573 ± 2% -42.9% 342823 ± 21% +29.4% 777397 ± 2% sched_debug.cpu.avg_idle.max
133080 ± 6% -48.5% 68563 ± 21% +87.9% 250058 ± 0% sched_debug.cpu.avg_idle.stddev
11.80 ± 22% +96.1% 23.13 ± 24% -60.4% 4.67 ± 2% sched_debug.cpu.clock.stddev
11.80 ± 22% +96.1% 23.13 ± 24% -60.4% 4.67 ± 2% sched_debug.cpu.clock_task.stddev
16.49 ± 3% +10.3% 18.19 ± 7% +2983.8% 508.41 ± 6% sched_debug.cpu.cpu_load[0].avg
38.35 ± 28% +69.4% 64.95 ± 61% +2275.1% 910.76 ± 1% sched_debug.cpu.cpu_load[0].max
7.67 ± 18% +53.8% 11.79 ± 53% +3832.5% 301.43 ± 5% sched_debug.cpu.cpu_load[0].stddev
16.39 ± 2% +9.9% 18.01 ± 5% +3723.3% 626.64 ± 3% sched_debug.cpu.cpu_load[1].avg
37.87 ± 27% +51.3% 57.30 ± 47% +2294.6% 906.91 ± 1% sched_debug.cpu.cpu_load[1].max
3.91 ± 17% +48.0% 5.78 ± 15% +4683.7% 187.00 ± 5% sched_debug.cpu.cpu_load[1].min
6.84 ± 20% +45.5% 9.95 ± 41% +2455.5% 174.75 ± 2% sched_debug.cpu.cpu_load[1].stddev
16.57 ± 2% +8.4% 17.96 ± 4% +3666.4% 624.20 ± 3% sched_debug.cpu.cpu_load[2].avg
37.71 ± 24% +38.8% 52.35 ± 36% +2301.1% 905.42 ± 1% sched_debug.cpu.cpu_load[2].max
6.02 ± 6% +18.8% 7.15 ± 8% +3322.5% 205.97 ± 2% sched_debug.cpu.cpu_load[2].min
6.50 ± 19% +36.7% 8.89 ± 31% +2513.9% 169.85 ± 1% sched_debug.cpu.cpu_load[2].stddev
16.99 ± 1% +6.4% 18.07 ± 3% +3565.7% 622.77 ± 3% sched_debug.cpu.cpu_load[3].avg
36.87 ± 19% +33.9% 49.39 ± 28% +2345.3% 901.64 ± 1% sched_debug.cpu.cpu_load[3].max
7.33 ± 3% +5.3% 7.72 ± 8% +2833.8% 214.97 ± 4% sched_debug.cpu.cpu_load[3].min
6.11 ± 15% +34.9% 8.24 ± 23% +2636.3% 167.13 ± 1% sched_debug.cpu.cpu_load[3].stddev
17.32 ± 1% +4.8% 18.15 ± 2% +3491.8% 622.26 ± 3% sched_debug.cpu.cpu_load[4].avg
35.56 ± 12% +32.8% 47.23 ± 22% +2414.9% 894.39 ± 1% sched_debug.cpu.cpu_load[4].max
8.00 ± 5% -2.0% 7.84 ± 8% +2683.7% 222.70 ± 6% sched_debug.cpu.cpu_load[4].min
5.80 ± 9% +35.4% 7.85 ± 18% +2705.5% 162.77 ± 0% sched_debug.cpu.cpu_load[4].stddev
16851 ± 1% -16.8% 14014 ± 3% -15.6% 14218 ± 3% sched_debug.cpu.curr->pid.avg
19325 ± 0% -19.1% 15644 ± 2% -6.4% 18083 ± 0% sched_debug.cpu.curr->pid.max
5114 ± 8% -48.9% 2611 ± 16% +20.8% 6179 ± 4% sched_debug.cpu.curr->pid.stddev
18.95 ± 7% -2.9% 18.40 ± 11% +3.7e+06% 708609 ± 3% sched_debug.cpu.load.avg
95.67 ± 46% -7.6% 88.42 ± 74% +1.1e+06% 1067053 ± 4% sched_debug.cpu.load.max
19.89 ± 31% -5.7% 18.76 ± 54% +1.7e+06% 338423 ± 3% sched_debug.cpu.load.stddev
500000 ± 0% +0.0% 500000 ± 0% +14.4% 572147 ± 4% sched_debug.cpu.max_idle_balance_cost.max
0.00 ± 4% +10.8% 0.00 ± 9% +35.0% 0.00 ± 12% sched_debug.cpu.next_balance.stddev
1417 ± 3% -6.8% 1322 ± 11% +31.2% 1860 ± 8% sched_debug.cpu.nr_load_updates.stddev
9.75 ± 5% -13.6% 8.43 ± 4% -7.0% 9.07 ± 11% sched_debug.cpu.nr_running.avg
29.22 ± 2% -16.8% 24.30 ± 7% +25.6% 36.70 ± 5% sched_debug.cpu.nr_running.max
7.47 ± 4% -20.3% 5.95 ± 5% +39.8% 10.44 ± 7% sched_debug.cpu.nr_running.stddev
10261512 ± 0% +132.6% 23873264 ± 3% -10.3% 9200003 ± 2% sched_debug.cpu.nr_switches.avg
11634045 ± 1% +126.0% 26295756 ± 2% -3.0% 11281317 ± 1% sched_debug.cpu.nr_switches.max
8958320 ± 1% +141.1% 21601624 ± 4% -15.9% 7538372 ± 3% sched_debug.cpu.nr_switches.min
780364 ± 7% +61.6% 1261065 ± 4% +65.0% 1287398 ± 2% sched_debug.cpu.nr_switches.stddev
8.65 ± 13% +23.4% 10.68 ± 13% +170.0% 23.36 ± 13% sched_debug.cpu.nr_uninterruptible.max
-13.62 ±-17% +45.2% -19.77 ±-23% +112.3% -28.91 ±-27% sched_debug.cpu.nr_uninterruptible.min
4.45 ± 8% +29.8% 5.78 ± 10% +106.3% 9.18 ± 24% sched_debug.cpu.nr_uninterruptible.stddev
Best Regards,
Huang, Ying
[toc] | [prev] | [next] | [standalone]
| From | Peter Zijlstra <peterz@infradead.org> |
|---|---|
| Date | 2016-06-01 10:50 +0200 |
| Subject | Re: [LKP] [lkp] [sched/fair] 53d3bc773e: hackbench.throughput -32.9% regression |
| Message-ID | <rFa0G-67A-23@gated-at.bofh.it> |
| In reply to | #1410820 |
On Wed, Jun 01, 2016 at 01:00:10PM +0800, Huang, Ying wrote: > Hi, Peter, > > Peter Zijlstra <peterz@infradead.org> writes: > > > On Tue, May 31, 2016 at 04:34:36PM +0800, Huang, Ying wrote: > >> Hi, Ingo, > >> > >> Part of the regression has been recovered in v4.7-rc1 from -32.9% to > >> -9.8%. But there is still some regression. Is it possible for fully > >> restore it? > > > > after much searching on how you guys run hackbench... I figured > > something like: > > > > perf bench sched messaging -g 20 --thread -l 60000 > > There is a reproduce file attached in the original report email, its > contents is something like below: > > 2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor <snip stupid large output> > 2016-05-15 09:06:24 /usr/bin/hackbench -g 24 --threads -l 60000 > > Hope that will help you for reproduce. It did not, because I didn't have the exact same machine and its not apparent how I should modify -- if at all -- the arguments to be representative when ran on my machine. > > on my IVB-EP (2*10*2) is similar to your IVT thing. > > > > And running something like: > > > > for i in /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor ; do echo performance > $i ; done > > perf stat --null --repeat 10 -- perf bench sched messaging -g 20 --thread -l 60000 | grep "seconds time elapsed" > > > > gets me: > > > > v4.6: > > > > 36.786914089 seconds time elapsed ( +- 0.49% ) > > 37.054017355 seconds time elapsed ( +- 1.05% ) > > > > > > origin/master (v4.7-rc1-ish): > > > > 34.757435264 seconds time elapsed ( +- 3.34% ) > > 35.396252515 seconds time elapsed ( +- 3.38% ) > > > > > > Which doesn't show a regression between v4.6 and HEAD; in fact it shows > > an improvement. > > Yes. For hackbench test, linus/master (v4.7-rc1+) is better than v4.6, > but it is worse than v4.6-rc7. Details is as below. That kernel was broken.. what your point?
[toc] | [prev] | [next] | [standalone]
| From | "Huang\, Ying" <ying.huang@intel.com> |
|---|---|
| Date | 2016-06-01 11:00 +0200 |
| Message-ID | <rFaam-6aY-15@gated-at.bofh.it> |
| In reply to | #1410971 |
Peter Zijlstra <peterz@infradead.org> writes: > On Wed, Jun 01, 2016 at 01:00:10PM +0800, Huang, Ying wrote: >> Hi, Peter, >> >> Peter Zijlstra <peterz@infradead.org> writes: >> >> > On Tue, May 31, 2016 at 04:34:36PM +0800, Huang, Ying wrote: >> >> Hi, Ingo, >> >> >> >> Part of the regression has been recovered in v4.7-rc1 from -32.9% to >> >> -9.8%. But there is still some regression. Is it possible for fully >> >> restore it? >> > >> > after much searching on how you guys run hackbench... I figured >> > something like: >> > >> > perf bench sched messaging -g 20 --thread -l 60000 >> >> There is a reproduce file attached in the original report email, its >> contents is something like below: >> >> 2016-05-15 08:57:02 echo performance > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor > > <snip stupid large output> > >> 2016-05-15 09:06:24 /usr/bin/hackbench -g 24 --threads -l 60000 >> >> Hope that will help you for reproduce. > > It did not, because I didn't have the exact same machine and its not > apparent how I should modify -- if at all -- the arguments to be > representative when ran on my machine. > >> > on my IVB-EP (2*10*2) is similar to your IVT thing. >> > >> > And running something like: >> > >> > for i in /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor ; do echo performance > $i ; done >> > perf stat --null --repeat 10 -- perf bench sched messaging -g 20 --thread -l 60000 | grep "seconds time elapsed" >> > >> > gets me: >> > >> > v4.6: >> > >> > 36.786914089 seconds time elapsed ( +- 0.49% ) >> > 37.054017355 seconds time elapsed ( +- 1.05% ) >> > >> > >> > origin/master (v4.7-rc1-ish): >> > >> > 34.757435264 seconds time elapsed ( +- 3.34% ) >> > 35.396252515 seconds time elapsed ( +- 3.38% ) >> > >> > >> > Which doesn't show a regression between v4.6 and HEAD; in fact it shows >> > an improvement. >> >> Yes. For hackbench test, linus/master (v4.7-rc1+) is better than v4.6, >> but it is worse than v4.6-rc7. Details is as below. > > That kernel was broken.. what your point? You mean the commit is a functionality fix? I found the hackbench.throughput for the test is v4.5: 1.4e+5 v4.6-rc1~v4.6-rc7: 1.9e+5 v4.6: 1.3e+5 v4.7-rc1: 1.7e+5 So some commit in v4.6-rc1 introduce some issue but improve the score for the test, which is fixed in v4.6, and some improvement merged by v4.7-rc1? Best Regards, Huang, Ying
[toc] | [prev] | [next] | [standalone]
| From | Mike Galbraith <umgwanakikbuti@gmail.com> |
|---|---|
| Date | 2016-06-01 11:50 +0200 |
| Subject | Re: [LKP] [lkp] [sched/fair] 53d3bc773e: hackbench.throughput -32.9% regression |
| Message-ID | <rFaWK-6IP-23@gated-at.bofh.it> |
| In reply to | #1410983 |
On Wed, 2016-06-01 at 16:53 +0800, Huang, Ying wrote: > You mean the commit is a functionality fix? Yup. These kind of things can be a bit annoying. The fix to not subtract load that we never added in the first place also pushed hackbench around a bit. Do anything at all with load_avg, you're gonna push this/that benchmark around a bit, and folks who watch numbers closely can end up chasing meaningless deltas. -Mike
[toc] | [prev] | [next] | [standalone]
| From | "Huang\, Ying" <ying.huang@intel.com> |
|---|---|
| Date | 2016-06-02 02:30 +0200 |
| Message-ID | <rFoGl-7g3-1@gated-at.bofh.it> |
| In reply to | #1411039 |
Mike Galbraith <umgwanakikbuti@gmail.com> writes: > On Wed, 2016-06-01 at 16:53 +0800, Huang, Ying wrote: > >> You mean the commit is a functionality fix? > > Yup. > > These kind of things can be a bit annoying. The fix to not subtract > load that we never added in the first place also pushed hackbench > around a bit. Do anything at all with load_avg, you're gonna push > this/that benchmark around a bit, and folks who watch numbers closely > can end up chasing meaningless deltas. If this report annoyed you, I am sorry about that. We just want to be helpful via providing some information. Now I know hackbench is sensitive to load_avg changes, thanks a lot for the information. Best Regards, Huang, Ying
[toc] | [prev] | [next] | [standalone]
| From | Mike Galbraith <umgwanakikbuti@gmail.com> |
|---|---|
| Date | 2016-06-02 02:50 +0200 |
| Subject | Re: [LKP] [lkp] [sched/fair] 53d3bc773e: hackbench.throughput -32.9% regression |
| Message-ID | <rFoZH-7mi-5@gated-at.bofh.it> |
| In reply to | #1411718 |
On Thu, 2016-06-02 at 08:28 +0800, Huang, Ying wrote: > If this report annoyed you, I am sorry about that. We just want to be > helpful via providing some information. Now I know hackbench is > sensitive to load_avg changes, thanks a lot for the information. No no, I meant the other way around ;-) If you're one of the folks _doing_ performance monitoring, it can be a tad annoying to finding that the effort you invested was all for naught. -Mike
[toc] | [prev] | [standalone]
Back to top | Article view | linux.kernel
csiph-web