* [sched] 143e1e28cb4: +17.9% aim7.jobs-per-min, -9.7% hackbench.throughput
@ 2014-08-10 4:41 Fengguang Wu
2014-08-10 7:59 ` Peter Zijlstra
0 siblings, 1 reply; 12+ messages in thread
From: Fengguang Wu @ 2014-08-10 4:41 UTC (permalink / raw)
To: Vincent Guittot
Cc: Dave Hansen, LKML, lkp, Ingo Molnar, Dietmar Eggemann,
Preeti U Murthy, Peter Zijlstra
Hi Vincent,
FYI, we noticed some performance ups/downs on
commit 143e1e28cb40bed836b0a06567208bd7347c9672 ("sched: Rework sched_domain topology definition")
107437febd495a5 143e1e28cb40bed836b0a0656 testbox/testcase/testparams
--------------- ------------------------- ---------------------------
0.09 ± 3% +88.2% 0.17 ± 1% nhm4/ebizzy/200%-100x-10s
0.09 ± 3% +88.2% 0.17 ± 1% TOTAL ebizzy.throughput.per_thread.stddev_percent
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
128529 ± 1% +17.9% 151594 ± 0% brickland1/aim7/6000-page_test
128529 ± 1% +17.9% 151594 ± 0% TOTAL aim7.jobs-per-min
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
156539 ± 1% -4.3% 149885 ± 0% lkp-snb01/hackbench/1600%-process-pipe
116465 ± 1% -17.1% 96542 ± 1% wsm/hackbench/1600%-process-pipe
273004 ± 1% -9.7% 246428 ± 0% TOTAL hackbench.throughput
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
12451869 ± 0% -2.9% 12087560 ± 0% brickland3/vm-scalability/300s-lru-file-readonce
12451869 ± 0% -2.9% 12087560 ± 0% TOTAL vm-scalability.throughput
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
1980 ± 0% -2.4% 1933 ± 0% nhm4/ebizzy/200%-100x-10s
1980 ± 0% -2.4% 1933 ± 0% TOTAL ebizzy.throughput.per_thread.min
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
4.446e+08 ± 0% -1.9% 4.364e+08 ± 0% lkp-nex04/pigz/100%-128K
4.446e+08 ± 0% -1.9% 4.364e+08 ± 0% TOTAL pigz.throughput
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
27.18 ± 0% -1.5% 26.78 ± 0% nhm4/ebizzy/200%-100x-10s
27.18 ± 0% -1.5% 26.78 ± 0% TOTAL ebizzy.time.user
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
2083 ± 0% +1.3% 2110 ± 0% nhm4/ebizzy/200%-100x-10s
2083 ± 0% +1.3% 2110 ± 0% TOTAL ebizzy.throughput.per_thread.max
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
32335 ± 0% -1.0% 32012 ± 0% nhm4/ebizzy/200%-100x-10s
32335 ± 0% -1.0% 32012 ± 0% TOTAL ebizzy.throughput
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
6626 ± 5% -88.7% 751 ±36% brickland3/vm-scalability/300s-lru-file-readonce
6626 ± 5% -88.7% 751 ±36% TOTAL cpuidle.C3-IVT.usage
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
14204402 ±19% -71.5% 4050186 ±34% brickland3/vm-scalability/300s-lru-file-readonce
14204402 ±19% -71.5% 4050186 ±34% TOTAL cpuidle.C3-IVT.time
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
0.38 ±32% +218.1% 1.20 ± 4% wsm/hackbench/1600%-process-pipe
0.38 ±32% +218.1% 1.20 ± 4% TOTAL perf-profile.cpu-cycles.__schedule.schedule.pipe_wait.pipe_read.do_sync_read
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
76064 ± 3% -32.2% 51572 ± 6% brickland1/aim7/6000-page_test
269053 ± 1% -57.9% 113386 ± 1% brickland3/vm-scalability/300s-lru-file-readonce
345117 ± 1% -52.2% 164959 ± 2% TOTAL cpuidle.C6-IVT.usage
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
59366697 ± 3% -46.1% 32017187 ± 7% brickland1/aim7/6000-page_test
59366697 ± 3% -46.1% 32017187 ± 7% TOTAL cpuidle.C1-IVT.time
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
26666815 ± 2% +83.3% 48893253 ± 2% lkp-nex04/pigz/100%-128K
26666815 ± 2% +83.3% 48893253 ± 2% TOTAL cpuidle.C1E-NHM.time
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
2561 ± 7% -42.9% 1463 ± 9% brickland1/aim7/6000-page_test
2561 ± 7% -42.9% 1463 ± 9% TOTAL numa-numastat.node2.other_node
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
116864 ± 2% +77.4% 207322 ± 2% lkp-nex04/pigz/100%-128K
116864 ± 2% +77.4% 207322 ± 2% TOTAL cpuidle.C1E-NHM.usage
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
0.65 ±22% +55.4% 1.02 ± 5% lkp-nex04/pigz/100%-128K
0.65 ±22% +55.4% 1.02 ± 5% TOTAL perf-profile.cpu-cycles.intel_idle.cpuidle_enter_state.cpuidle_enter.cpu_startup_entry.start_secondary
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
9926 ± 2% -43.8% 5577 ± 4% brickland1/aim7/6000-page_test
9926 ± 2% -43.8% 5577 ± 4% TOTAL proc-vmstat.numa_other
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
2627 ±12% -49.1% 1337 ±12% brickland1/aim7/6000-page_test
2627 ±12% -49.1% 1337 ±12% TOTAL numa-numastat.node1.other_node
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
0.88 ± 6% +65.8% 1.46 ± 4% lkp-nex04/pigz/100%-128K
0.88 ± 6% +65.8% 1.46 ± 4% TOTAL turbostat.%c3
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
19542 ± 9% -38.3% 12057 ± 4% brickland1/aim7/6000-page_test
19542 ± 9% -38.3% 12057 ± 4% TOTAL cpuidle.C1E-IVT.usage
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
1357 ± 8% -39.0% 827 ±16% brickland3/vm-scalability/300s-lru-file-readonce
1357 ± 8% -39.0% 827 ±16% TOTAL slabinfo.nfs_write_data.num_objs
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
1357 ± 8% -39.0% 827 ±16% brickland3/vm-scalability/300s-lru-file-readonce
1357 ± 8% -39.0% 827 ±16% TOTAL slabinfo.nfs_write_data.active_objs
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
993654 ± 2% -19.9% 795962 ± 3% brickland1/aim7/6000-page_test
343162 ± 6% -59.1% 140462 ± 5% brickland3/vm-scalability/300s-lru-file-readonce
315034 ± 3% +34.5% 423784 ± 1% wsm/hackbench/1600%-process-pipe
1651850 ± 3% -17.7% 1360209 ± 2% TOTAL softirqs.RCU
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
2455 ±10% -41.0% 1448 ± 9% brickland1/aim7/6000-page_test
2455 ±10% -41.0% 1448 ± 9% TOTAL numa-numastat.node0.other_node
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
582269 ±14% -55.6% 258617 ±16% brickland1/aim7/6000-page_test
441740 ±18% -49.2% 224389 ± 0% brickland3/vm-scalability/300s-lru-file-readonce
219830 ± 1% -29.9% 154078 ± 0% lkp-nex04/pigz/100%-128K
156140 ± 8% -12.1% 137250 ± 5% lkp-snb01/hackbench/1600%-process-pipe
47877 ± 1% -10.3% 42941 ± 1% wsm/hackbench/1600%-process-pipe
1447857 ±12% -43.6% 817277 ± 6% TOTAL softirqs.SCHED
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
892836 ± 1% +58.0% 1410897 ± 0% lkp-nex04/pigz/100%-128K
892836 ± 1% +58.0% 1410897 ± 0% TOTAL cpuidle.C3-NHM.usage
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
471304 ±11% -31.4% 323251 ± 8% brickland1/aim7/6000-page_test
471304 ±11% -31.4% 323251 ± 8% TOTAL numa-vmstat.node1.nr_anon_pages
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
4.875e+08 ± 2% +57.7% 7.688e+08 ± 2% lkp-nex04/pigz/100%-128K
4.875e+08 ± 2% +57.7% 7.688e+08 ± 2% TOTAL cpuidle.C3-NHM.time
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
2281 ±12% -41.8% 1327 ±16% brickland1/aim7/6000-page_test
2281 ±12% -41.8% 1327 ±16% TOTAL numa-numastat.node3.other_node
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
1903446 ±11% -30.7% 1318156 ± 7% brickland1/aim7/6000-page_test
1903446 ±11% -30.7% 1318156 ± 7% TOTAL numa-meminfo.node1.AnonPages
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
1.64 ± 3% -44.8% 0.90 ± 4% brickland3/vm-scalability/300s-lru-file-readonce
1.83 ± 1% +54.8% 2.84 ± 0% lkp-nex04/pigz/100%-128K
1.26 ± 2% -21.9% 0.99 ± 6% wsm/hackbench/1600%-process-pipe
4.73 ± 2% -0.1% 4.73 ± 2% TOTAL turbostat.%c1
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
518274 ±11% -30.4% 360742 ± 8% brickland1/aim7/6000-page_test
518274 ±11% -30.4% 360742 ± 8% TOTAL numa-vmstat.node1.nr_active_anon
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
2097138 ±10% -30.0% 1469003 ± 8% brickland1/aim7/6000-page_test
2097138 ±10% -30.0% 1469003 ± 8% TOTAL numa-meminfo.node1.Active(anon)
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
49527464 ± 6% -32.4% 33488833 ± 4% brickland1/aim7/6000-page_test
49527464 ± 6% -32.4% 33488833 ± 4% TOTAL cpuidle.C1E-IVT.time
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
2084 ±10% +31.6% 2743 ± 6% lkp-snb01/hackbench/1600%-process-pipe
235 ± 8% -35.7% 151 ± 8% wsm/hackbench/1600%-process-pipe
2319 ±10% +24.8% 2894 ± 6% TOTAL cpuidle.POLL.usage
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
54543 ±11% -37.2% 34252 ±16% brickland1/aim7/6000-page_test
4542 ± 4% -14.3% 3891 ± 6% brickland3/vm-scalability/300s-lru-file-readonce
59085 ±10% -35.4% 38143 ±15% TOTAL cpuidle.C1-IVT.usage
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
188938 ±33% -41.3% 110966 ±16% brickland1/aim7/6000-page_test
188938 ±33% -41.3% 110966 ±16% TOTAL numa-meminfo.node2.PageTables
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
47262 ±35% -42.3% 27273 ±16% brickland1/aim7/6000-page_test
47262 ±35% -42.3% 27273 ±16% TOTAL numa-vmstat.node2.nr_page_table_pages
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
1944687 ±10% -25.8% 1443923 ±16% brickland1/aim7/6000-page_test
1944687 ±10% -25.8% 1443923 ±16% TOTAL numa-meminfo.node3.Active(anon)
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
1754763 ±11% -26.6% 1288713 ±16% brickland1/aim7/6000-page_test
1754763 ±11% -26.6% 1288713 ±16% TOTAL numa-meminfo.node3.AnonPages
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
1964722 ±10% -25.5% 1464696 ±16% brickland1/aim7/6000-page_test
1964722 ±10% -25.5% 1464696 ±16% TOTAL numa-meminfo.node3.Active
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
17100655 ± 7% +44.6% 24731604 ± 2% lkp-nex04/pigz/100%-128K
37335977 ± 6% -28.6% 26654124 ± 7% wsm/hackbench/1600%-process-pipe
54436632 ± 6% -5.6% 51385728 ± 5% TOTAL cpuidle.C1-NHM.time
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
432109 ± 9% -26.2% 318886 ±14% brickland1/aim7/6000-page_test
432109 ± 9% -26.2% 318886 ±14% TOTAL numa-vmstat.node3.nr_anon_pages
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
479527 ± 9% -25.3% 358029 ±14% brickland1/aim7/6000-page_test
479527 ± 9% -25.3% 358029 ±14% TOTAL numa-vmstat.node3.nr_active_anon
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
3157742 ±16% -26.5% 2320253 ±10% brickland1/aim7/6000-page_test
3157742 ±16% -26.5% 2320253 ±10% TOTAL numa-meminfo.node1.MemUsed
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
2.00 ±39% -34.4% 1.31 ±10% lkp-nex04/pigz/100%-128K
2.00 ±39% -34.4% 1.31 ±10% TOTAL perf-profile.cpu-cycles.update_cfs_shares.task_tick_fair.scheduler_tick.update_process_times.tick_sched_handle
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
9869 ±15% +21.8% 12025 ± 1% lkp-snb01/hackbench/1600%-process-pipe
9869 ±15% +21.8% 12025 ± 1% TOTAL numa-vmstat.node1.numa_other
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
61500 ± 2% +24.6% 76639 ± 2% lkp-nex04/pigz/100%-128K
1206181 ± 3% -30.5% 838619 ± 2% wsm/hackbench/1600%-process-pipe
1267682 ± 3% -27.8% 915259 ± 2% TOTAL cpuidle.C1-NHM.usage
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
2118206 ±10% -29.7% 1488874 ± 7% brickland1/aim7/6000-page_test
593077 ± 4% -9.9% 534490 ± 1% lkp-snb01/hackbench/1600%-process-pipe
2711283 ± 9% -25.4% 2023365 ± 6% TOTAL numa-meminfo.node1.Active
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
227935 ±13% -30.2% 159051 ±16% brickland3/vm-scalability/300s-lru-file-readonce
227935 ±13% -30.2% 159051 ±16% TOTAL numa-vmstat.node0.workingset_nodereclaim
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
7303589 ± 2% -24.8% 5495829 ± 3% brickland1/aim7/6000-page_test
202793 ± 3% +20.0% 243307 ± 2% wsm/hackbench/1600%-process-pipe
7506382 ± 2% -23.5% 5739136 ± 3% TOTAL meminfo.AnonPages
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
2064106 ± 7% -23.3% 1582792 ± 8% brickland1/aim7/6000-page_test
2064106 ± 7% -23.3% 1582792 ± 8% TOTAL numa-meminfo.node0.Active
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
8189 ± 9% -18.6% 6669 ± 3% lkp-snb01/hackbench/1600%-process-pipe
8189 ± 9% -18.6% 6669 ± 3% TOTAL slabinfo.proc_inode_cache.active_objs
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
8064024 ± 2% -24.0% 6132677 ± 3% brickland1/aim7/6000-page_test
203300 ± 3% +19.6% 243236 ± 2% wsm/hackbench/1600%-process-pipe
8267324 ± 2% -22.9% 6375914 ± 3% TOTAL meminfo.Active(anon)
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
2815 ± 3% -8.3% 2581 ± 6% brickland3/vm-scalability/300s-lru-file-readonce
1076 ±15% +19.6% 1287 ±12% nhm4/ebizzy/200%-100x-10s
3892 ± 6% -0.6% 3868 ± 8% TOTAL slabinfo.buffer_head.num_objs
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
6.567e+11 ± 3% -21.4% 5.16e+11 ± 4% brickland1/aim7/6000-page_test
1872397 ± 3% +22.2% 2288513 ± 2% wsm/hackbench/1600%-process-pipe
6.567e+11 ± 3% -21.4% 5.16e+11 ± 4% TOTAL meminfo.Committed_AS
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
235358 ± 5% -19.8% 188793 ± 3% brickland1/aim7/6000-page_test
235358 ± 5% -19.8% 188793 ± 3% TOTAL proc-vmstat.pgmigrate_success
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
235358 ± 5% -19.8% 188793 ± 3% brickland1/aim7/6000-page_test
235358 ± 5% -19.8% 188793 ± 3% TOTAL proc-vmstat.numa_pages_migrated
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
433235 ± 4% -18.1% 354845 ± 5% brickland1/aim7/6000-page_test
433235 ± 4% -18.1% 354845 ± 5% TOTAL numa-vmstat.node2.nr_anon_pages
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
90514624 ± 2% -21.6% 70928192 ± 1% wsm/hackbench/1600%-process-pipe
90514624 ± 2% -21.6% 70928192 ± 1% TOTAL proc-vmstat.pgalloc_dma32
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
463719 ± 8% -24.7% 349388 ± 7% brickland1/aim7/6000-page_test
71979 ± 2% -11.7% 63550 ± 4% lkp-snb01/hackbench/1600%-process-pipe
535698 ± 7% -22.9% 412938 ± 7% TOTAL numa-vmstat.node0.nr_anon_pages
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
1818612 ± 2% -24.9% 1365670 ± 3% brickland1/aim7/6000-page_test
51594 ± 2% +16.9% 60313 ± 1% wsm/hackbench/1600%-process-pipe
1870207 ± 2% -23.8% 1425983 ± 3% TOTAL proc-vmstat.nr_anon_pages
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
3187 ± 5% -18.5% 2599 ± 6% brickland1/aim7/6000-page_test
3187 ± 5% -18.5% 2599 ± 6% TOTAL numa-vmstat.node0.numa_other
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
2007155 ± 2% -24.3% 1518688 ± 3% brickland1/aim7/6000-page_test
51734 ± 2% +16.6% 60298 ± 1% wsm/hackbench/1600%-process-pipe
2058889 ± 2% -23.3% 1578987 ± 3% TOTAL proc-vmstat.nr_active_anon
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
1395062 ± 6% -19.0% 1130108 ± 3% brickland1/aim7/6000-page_test
1395062 ± 6% -19.0% 1130108 ± 3% TOTAL proc-vmstat.numa_hint_faults
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
2501 ± 9% +25.0% 3126 ±11% brickland3/vm-scalability/300s-lru-file-readonce
2501 ± 9% +25.0% 3126 ±11% TOTAL proc-vmstat.compact_stall
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
477037 ± 4% -17.2% 394983 ± 5% brickland1/aim7/6000-page_test
477037 ± 4% -17.2% 394983 ± 5% TOTAL numa-vmstat.node2.nr_active_anon
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
511455 ± 8% -23.9% 389447 ± 7% brickland1/aim7/6000-page_test
71970 ± 2% -11.7% 63552 ± 4% lkp-snb01/hackbench/1600%-process-pipe
583426 ± 7% -22.4% 452999 ± 7% TOTAL numa-vmstat.node0.nr_active_anon
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
796281 ±23% -27.7% 575352 ± 3% brickland1/aim7/6000-page_test
157314 ± 2% +22.7% 193026 ± 2% wsm/hackbench/1600%-process-pipe
953596 ±19% -19.4% 768378 ± 3% TOTAL meminfo.PageTables
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
2829 ±10% +18.7% 3357 ± 3% brickland1/aim7/6000-page_test
2829 ±10% +18.7% 3357 ± 3% TOTAL numa-vmstat.node2.nr_alloc_batch
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
1850230 ± 8% -24.1% 1405061 ± 8% brickland1/aim7/6000-page_test
289636 ± 2% -12.3% 254041 ± 4% lkp-snb01/hackbench/1600%-process-pipe
2139866 ± 7% -22.5% 1659103 ± 8% TOTAL numa-meminfo.node0.AnonPages
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
8145316 ± 2% -23.7% 6213832 ± 3% brickland1/aim7/6000-page_test
353534 ± 2% +11.3% 393604 ± 1% wsm/hackbench/1600%-process-pipe
8498851 ± 2% -22.3% 6607437 ± 3% TOTAL meminfo.Active
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
2706 ± 4% +26.1% 3411 ± 5% brickland1/aim7/6000-page_test
2706 ± 4% +26.1% 3411 ± 5% TOTAL numa-vmstat.node1.nr_alloc_batch
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
1.392e+08 ± 5% -15.5% 1.176e+08 ± 2% lkp-snb01/hackbench/1600%-process-pipe
1.392e+08 ± 5% -15.5% 1.176e+08 ± 2% TOTAL numa-numastat.node1.local_node
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
1.392e+08 ± 5% -15.5% 1.176e+08 ± 2% lkp-snb01/hackbench/1600%-process-pipe
1.392e+08 ± 5% -15.5% 1.176e+08 ± 2% TOTAL numa-numastat.node1.numa_hit
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
2044097 ± 7% -23.5% 1562809 ± 8% brickland1/aim7/6000-page_test
289613 ± 2% -12.3% 254058 ± 4% lkp-snb01/hackbench/1600%-process-pipe
2333711 ± 7% -22.1% 1816867 ± 7% TOTAL numa-meminfo.node0.Active(anon)
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
198747 ±23% -28.0% 143034 ± 3% brickland1/aim7/6000-page_test
40081 ± 2% +19.1% 47725 ± 1% wsm/hackbench/1600%-process-pipe
238828 ±19% -20.1% 190760 ± 2% TOTAL proc-vmstat.nr_page_table_pages
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
1.65e+08 ± 2% -11.3% 1.463e+08 ± 2% lkp-snb01/hackbench/1600%-process-pipe
82216748 ± 2% -23.6% 62851371 ± 1% wsm/hackbench/1600%-process-pipe
2.472e+08 ± 2% -15.4% 2.092e+08 ± 2% TOTAL proc-vmstat.pgalloc_normal
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
2725835 ± 4% -17.5% 2247537 ± 4% brickland1/aim7/6000-page_test
2725835 ± 4% -17.5% 2247537 ± 4% TOTAL numa-meminfo.node2.MemUsed
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
393637 ± 6% -15.3% 333296 ± 2% brickland1/aim7/6000-page_test
393637 ± 6% -15.3% 333296 ± 2% TOTAL proc-vmstat.numa_hint_faults_local
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
1.681e+08 ± 2% -10.6% 1.504e+08 ± 3% lkp-snb01/hackbench/1600%-process-pipe
1.709e+08 ± 2% -22.6% 1.323e+08 ± 1% wsm/hackbench/1600%-process-pipe
3.391e+08 ± 2% -16.6% 2.827e+08 ± 2% TOTAL proc-vmstat.numa_local
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
1.681e+08 ± 2% -10.6% 1.504e+08 ± 3% lkp-snb01/hackbench/1600%-process-pipe
1.709e+08 ± 2% -22.6% 1.323e+08 ± 1% wsm/hackbench/1600%-process-pipe
3.391e+08 ± 2% -16.6% 2.827e+08 ± 2% TOTAL proc-vmstat.numa_hit
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
109532 ± 1% +20.5% 132015 ± 2% wsm/hackbench/1600%-process-pipe
109532 ± 1% +20.5% 132015 ± 2% TOTAL slabinfo.vm_area_struct.active_objs
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
1.706e+08 ± 2% -10.5% 1.527e+08 ± 2% lkp-snb01/hackbench/1600%-process-pipe
1.727e+08 ± 2% -22.6% 1.338e+08 ± 1% wsm/hackbench/1600%-process-pipe
3.433e+08 ± 2% -16.5% 2.865e+08 ± 2% TOTAL proc-vmstat.pgfree
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
2.82 ± 3% +21.9% 3.43 ± 4% brickland1/aim7/6000-page_test
2.82 ± 3% +21.9% 3.43 ± 4% TOTAL turbostat.%pc2
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
1045052 ± 1% -15.9% 878536 ± 3% brickland3/vm-scalability/300s-lru-file-readonce
1045052 ± 1% -15.9% 878536 ± 3% TOTAL proc-vmstat.workingset_nodereclaim
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
148761 ± 1% +19.9% 178354 ± 2% wsm/hackbench/1600%-process-pipe
148761 ± 1% +19.9% 178354 ± 2% TOTAL slabinfo.kmalloc-64.active_objs
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
3023001 ± 1% -15.1% 2565273 ± 3% brickland3/vm-scalability/300s-lru-file-readonce
3023001 ± 1% -15.1% 2565273 ± 3% TOTAL proc-vmstat.slabs_scanned
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
74702 ± 1% +19.2% 89026 ± 2% wsm/hackbench/1600%-process-pipe
74702 ± 1% +19.2% 89026 ± 2% TOTAL slabinfo.anon_vma.active_objs
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
14041931 ± 2% -16.0% 11788641 ± 1% wsm/hackbench/1600%-process-pipe
14041931 ± 2% -16.0% 11788641 ± 1% TOTAL proc-vmstat.pgfault
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
1742111 ± 4% -16.9% 1447181 ± 5% brickland1/aim7/6000-page_test
1742111 ± 4% -16.9% 1447181 ± 5% TOTAL numa-meminfo.node2.AnonPages
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
15865125 ± 1% -15.0% 13485882 ± 1% brickland1/aim7/6000-page_test
15865125 ± 1% -15.0% 13485882 ± 1% TOTAL softirqs.TIMER
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
1923000 ± 4% -16.4% 1608509 ± 5% brickland1/aim7/6000-page_test
1923000 ± 4% -16.4% 1608509 ± 5% TOTAL numa-meminfo.node2.Active(anon)
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
1943185 ± 4% -16.2% 1629057 ± 5% brickland1/aim7/6000-page_test
1943185 ± 4% -16.2% 1629057 ± 5% TOTAL numa-meminfo.node2.Active
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
1.659e+08 ± 2% -13.8% 1.429e+08 ± 1% wsm/hackbench/1600%-process-pipe
1.659e+08 ± 2% -13.8% 1.429e+08 ± 1% TOTAL cpuidle.C6-NHM.time
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
2744 ± 1% +16.7% 3202 ± 2% wsm/hackbench/1600%-process-pipe
2744 ± 1% +16.7% 3202 ± 2% TOTAL slabinfo.vm_area_struct.active_slabs
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
2744 ± 1% +16.7% 3202 ± 2% wsm/hackbench/1600%-process-pipe
2744 ± 1% +16.7% 3202 ± 2% TOTAL slabinfo.vm_area_struct.num_slabs
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
120778 ± 1% +16.7% 140916 ± 2% wsm/hackbench/1600%-process-pipe
120778 ± 1% +16.7% 140916 ± 2% TOTAL slabinfo.vm_area_struct.num_objs
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
2107 ± 7% +14.9% 2420 ± 8% brickland3/vm-scalability/300s-lru-file-readonce
2107 ± 7% +14.9% 2420 ± 8% TOTAL proc-vmstat.compact_success
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
2591 ± 1% +16.0% 3007 ± 2% wsm/hackbench/1600%-process-pipe
2591 ± 1% +16.0% 3007 ± 2% TOTAL slabinfo.kmalloc-64.active_slabs
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
2591 ± 1% +16.0% 3007 ± 2% wsm/hackbench/1600%-process-pipe
2591 ± 1% +16.0% 3007 ± 2% TOTAL slabinfo.kmalloc-64.num_slabs
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
165897 ± 1% +16.0% 192484 ± 2% wsm/hackbench/1600%-process-pipe
165897 ± 1% +16.0% 192484 ± 2% TOTAL slabinfo.kmalloc-64.num_objs
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
4.40 ± 2% +22.0% 5.37 ± 4% brickland1/aim7/6000-page_test
1.54 ± 2% -12.0% 1.35 ± 3% wsm/hackbench/1600%-process-pipe
5.94 ± 2% +13.2% 6.73 ± 4% TOTAL turbostat.%c6
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
1683 ± 5% -11.0% 1497 ± 9% brickland3/vm-scalability/300s-lru-file-readonce
1683 ± 5% -11.0% 1497 ± 9% TOTAL slabinfo.xfs_inode.num_objs
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
1683 ± 5% -11.0% 1497 ± 9% brickland3/vm-scalability/300s-lru-file-readonce
1683 ± 5% -11.0% 1497 ± 9% TOTAL slabinfo.xfs_inode.active_objs
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
329 ± 1% -13.3% 285 ± 0% brickland1/aim7/6000-page_test
329 ± 1% -13.3% 285 ± 0% TOTAL uptime.boot
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
43214 ± 6% -15.2% 36648 ± 4% lkp-snb01/hackbench/1600%-process-pipe
43214 ± 6% -15.2% 36648 ± 4% TOTAL numa-vmstat.node0.nr_page_table_pages
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
175030 ± 7% -16.1% 146779 ± 4% lkp-snb01/hackbench/1600%-process-pipe
175030 ± 7% -16.1% 146779 ± 4% TOTAL numa-meminfo.node0.PageTables
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
47374 ± 1% -10.4% 42459 ± 4% wsm/hackbench/1600%-process-pipe
47374 ± 1% -10.4% 42459 ± 4% TOTAL meminfo.DirectMap4k
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
205471 ± 7% +18.5% 243577 ± 3% lkp-snb01/hackbench/1600%-process-pipe
205471 ± 7% +18.5% 243577 ± 3% TOTAL cpuidle.C1E-SNB.usage
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
86033 ± 1% +14.0% 98061 ± 2% wsm/hackbench/1600%-process-pipe
86033 ± 1% +14.0% 98061 ± 2% TOTAL slabinfo.anon_vma.num_objs
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
10692 ± 1% -12.7% 9334 ± 3% brickland3/vm-scalability/300s-lru-file-readonce
10692 ± 1% -12.7% 9334 ± 3% TOTAL proc-vmstat.pageoutrun
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
1343 ± 1% +14.0% 1531 ± 2% wsm/hackbench/1600%-process-pipe
1343 ± 1% +14.0% 1531 ± 2% TOTAL slabinfo.anon_vma.active_slabs
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
1343 ± 1% +14.0% 1531 ± 2% wsm/hackbench/1600%-process-pipe
1343 ± 1% +14.0% 1531 ± 2% TOTAL slabinfo.anon_vma.num_slabs
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
18554 ± 0% -9.2% 16845 ± 0% lkp-snb01/hackbench/1600%-process-pipe
4687 ± 1% +19.1% 5582 ± 2% wsm/hackbench/1600%-process-pipe
23241 ± 0% -3.5% 22427 ± 1% TOTAL slabinfo.mm_struct.active_objs
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
8281 ± 0% -14.7% 7062 ± 2% lkp-snb01/hackbench/1600%-process-pipe
3737 ± 0% -9.8% 3371 ± 1% wsm/hackbench/1600%-process-pipe
12018 ± 0% -13.2% 10433 ± 1% TOTAL vmstat.procs.r
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
26179 ± 0% -10.1% 23527 ± 0% lkp-snb01/hackbench/1600%-process-pipe
5556 ± 1% +16.6% 6479 ± 2% wsm/hackbench/1600%-process-pipe
31736 ± 0% -5.4% 30006 ± 1% TOTAL slabinfo.files_cache.active_objs
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
3077 ± 1% +14.5% 3523 ± 0% brickland1/aim7/6000-page_test
181562 ± 3% +13.5% 206096 ± 5% brickland3/vm-scalability/300s-lru-file-readonce
184639 ± 3% +13.5% 209619 ± 5% TOTAL proc-vmstat.pgactivate
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
6012 ± 1% +13.1% 6797 ± 1% wsm/hackbench/1600%-process-pipe
6012 ± 1% +13.1% 6797 ± 1% TOTAL slabinfo.mm_struct.num_objs
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
13158 ±13% -14.4% 11261 ± 4% brickland1/aim7/6000-page_test
13158 ±13% -14.4% 11261 ± 4% TOTAL numa-meminfo.node3.SReclaimable
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
3289 ±13% -14.4% 2815 ± 4% brickland1/aim7/6000-page_test
3289 ±13% -14.4% 2815 ± 4% TOTAL numa-vmstat.node3.nr_slab_reclaimable
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
6722 ± 1% +12.5% 7562 ± 2% wsm/hackbench/1600%-process-pipe
6722 ± 1% +12.5% 7562 ± 2% TOTAL slabinfo.files_cache.num_objs
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
9663 ± 0% -11.6% 8546 ± 2% brickland3/vm-scalability/300s-lru-file-readonce
9663 ± 0% -11.6% 8546 ± 2% TOTAL proc-vmstat.kswapd_low_wmark_hit_quickly
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
60059 ± 5% +12.2% 67416 ± 3% lkp-snb01/hackbench/1600%-process-pipe
60059 ± 5% +12.2% 67416 ± 3% TOTAL cpuidle.C7-SNB.usage
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
2815 ± 3% -8.3% 2581 ± 6% brickland3/vm-scalability/300s-lru-file-readonce
2815 ± 3% -8.3% 2581 ± 6% TOTAL slabinfo.buffer_head.active_objs
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
4529818 ± 1% -8.8% 4129398 ± 1% brickland1/aim7/6000-page_test
237630 ± 0% -13.1% 206405 ± 0% brickland3/vm-scalability/300s-lru-file-readonce
8.785e+08 ± 1% +51.3% 1.329e+09 ± 1% lkp-snb01/hackbench/1600%-process-pipe
89463167 ± 9% +250.9% 3.14e+08 ± 1% wsm/hackbench/1600%-process-pipe
9.727e+08 ± 2% +69.3% 1.647e+09 ± 1% TOTAL time.involuntary_context_switches
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
2175 ± 1% -8.8% 1984 ± 0% brickland3/vm-scalability/300s-lru-file-readonce
23054 ± 1% +12.2% 25876 ± 0% lkp-nex04/pigz/100%-128K
4619649 ± 0% +30.1% 6008008 ± 2% lkp-snb01/hackbench/1600%-process-pipe
529899 ± 5% +167.6% 1418080 ± 0% wsm/hackbench/1600%-process-pipe
5174778 ± 1% +44.0% 7453950 ± 2% TOTAL vmstat.system.cs
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
3150464 ± 2% -24.2% 2387551 ± 3% brickland1/aim7/6000-page_test
3890218 ± 1% +7.6% 4184429 ± 0% lkp-nex04/pigz/100%-128K
1.926e+09 ± 1% +21.0% 2.331e+09 ± 1% lkp-snb01/hackbench/1600%-process-pipe
2.322e+08 ± 3% +136.4% 5.489e+08 ± 0% wsm/hackbench/1600%-process-pipe
2.166e+09 ± 1% +33.3% 2.887e+09 ± 1% TOTAL time.voluntary_context_switches
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
1067968 ± 1% -36.5% 677985 ± 4% lkp-snb01/hackbench/1600%-process-pipe
231933 ± 0% -1.2% 229082 ± 0% nhm4/ebizzy/200%-100x-10s
16014 ± 1% +59.3% 25511 ± 0% wsm/hackbench/1600%-process-pipe
1315917 ± 0% -29.1% 932579 ± 2% TOTAL vmstat.system.in
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
281 ± 1% -15.1% 238 ± 0% brickland1/aim7/6000-page_test
281 ± 1% -15.1% 238 ± 0% TOTAL time.elapsed_time
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
29294 ± 1% -14.3% 25093 ± 0% brickland1/aim7/6000-page_test
33619 ± 0% +1.2% 34032 ± 0% brickland3/vm-scalability/300s-lru-file-readonce
62914 ± 0% -6.0% 59125 ± 0% TOTAL time.system_time
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
18258004 ± 1% -3.8% 17557683 ± 1% lkp-snb01/hackbench/1600%-process-pipe
3.104e+09 ± 0% -1.0% 3.073e+09 ± 0% nhm4/ebizzy/200%-100x-10s
13852545 ± 2% -16.7% 11538015 ± 1% wsm/hackbench/1600%-process-pipe
3.137e+09 ± 0% -1.1% 3.103e+09 ± 0% TOTAL time.minor_page_faults
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
18531 ± 0% -1.8% 18206 ± 0% lkp-nex04/pigz/100%-128K
1388 ± 0% +11.2% 1543 ± 0% lkp-snb01/hackbench/1600%-process-pipe
2718 ± 0% -1.5% 2678 ± 0% nhm4/ebizzy/200%-100x-10s
22638 ± 0% -0.9% 22428 ± 0% TOTAL time.user_time
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
± 1% -3.4% ± 0% brickland1/aim7/6000-page_test
± 1% -3.4% ± 0% TOTAL turbostat.RAM_W
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
± 0% +1.8% ± 0% lkp-snb01/hackbench/1600%-process-pipe
± 0% +1.8% ± 0% TOTAL turbostat.Cor_W
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
10655 ± 0% +1.4% 10802 ± 0% brickland1/aim7/6000-page_test
8224 ± 0% +1.4% 8341 ± 0% brickland3/vm-scalability/300s-lru-file-readonce
6230 ± 0% -1.7% 6125 ± 0% lkp-nex04/pigz/100%-128K
25110 ± 0% +0.6% 25269 ± 0% TOTAL time.percent_of_cpu_this_job_got
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
72.35 ± 0% +1.2% 73.20 ± 0% brickland3/vm-scalability/300s-lru-file-readonce
97.29 ± 0% -1.6% 95.71 ± 0% lkp-nex04/pigz/100%-128K
169.63 ± 0% -0.4% 168.91 ± 0% TOTAL turbostat.%c0
107437febd495a5 143e1e28cb40bed836b0a0656
--------------- -------------------------
± 0% +1.2% ± 0% lkp-snb01/hackbench/1600%-process-pipe
± 0% +1.2% ± 0% TOTAL turbostat.Pkg_W
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,
Fengguang
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [sched] 143e1e28cb4: +17.9% aim7.jobs-per-min, -9.7% hackbench.throughput
2014-08-10 4:41 [sched] 143e1e28cb4: +17.9% aim7.jobs-per-min, -9.7% hackbench.throughput Fengguang Wu
@ 2014-08-10 7:59 ` Peter Zijlstra
2014-08-10 10:54 ` Fengguang Wu
0 siblings, 1 reply; 12+ messages in thread
From: Peter Zijlstra @ 2014-08-10 7:59 UTC (permalink / raw)
To: Fengguang Wu
Cc: Vincent Guittot, Dave Hansen, LKML, lkp, Ingo Molnar,
Dietmar Eggemann, Preeti U Murthy
[-- Attachment #1: Type: text/plain, Size: 862 bytes --]
On Sun, Aug 10, 2014 at 12:41:27PM +0800, Fengguang Wu wrote:
> Hi Vincent,
>
> FYI, we noticed some performance ups/downs on
>
> commit 143e1e28cb40bed836b0a06567208bd7347c9672 ("sched: Rework sched_domain topology definition")
>
> 128529 ± 1% +17.9% 151594 ± 0% brickland1/aim7/6000-page_test
> 76064 ± 3% -32.2% 51572 ± 6% brickland1/aim7/6000-page_test
> 59366697 ± 3% -46.1% 32017187 ± 7% brickland1/aim7/6000-page_test
> 2561 ± 7% -42.9% 1463 ± 9% brickland1/aim7/6000-page_test
> 9926 ± 2% -43.8% 5577 ± 4% brickland1/aim7/6000-page_test
> 19542 ± 9% -38.3% 12057 ± 4% brickland1/aim7/6000-page_test
> 993654 ± 2% -19.9% 795962 ± 3% brickland1/aim7/6000-page_test
etc..
how does one read that? afaict its a random number generator..
[-- Attachment #2: Type: application/pgp-signature, Size: 836 bytes --]
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [sched] 143e1e28cb4: +17.9% aim7.jobs-per-min, -9.7% hackbench.throughput
2014-08-10 7:59 ` Peter Zijlstra
@ 2014-08-10 10:54 ` Fengguang Wu
2014-08-10 15:05 ` Peter Zijlstra
2014-08-11 13:33 ` Peter Zijlstra
0 siblings, 2 replies; 12+ messages in thread
From: Fengguang Wu @ 2014-08-10 10:54 UTC (permalink / raw)
To: Peter Zijlstra
Cc: Vincent Guittot, Dave Hansen, LKML, lkp, Ingo Molnar,
Dietmar Eggemann, Preeti U Murthy
On Sun, Aug 10, 2014 at 09:59:15AM +0200, Peter Zijlstra wrote:
> On Sun, Aug 10, 2014 at 12:41:27PM +0800, Fengguang Wu wrote:
> > Hi Vincent,
> >
> > FYI, we noticed some performance ups/downs on
> >
> > commit 143e1e28cb40bed836b0a06567208bd7347c9672 ("sched: Rework sched_domain topology definition")
> >
> > 128529 ± 1% +17.9% 151594 ± 0% brickland1/aim7/6000-page_test
> > 76064 ± 3% -32.2% 51572 ± 6% brickland1/aim7/6000-page_test
> > 59366697 ± 3% -46.1% 32017187 ± 7% brickland1/aim7/6000-page_test
> > 2561 ± 7% -42.9% 1463 ± 9% brickland1/aim7/6000-page_test
> > 9926 ± 2% -43.8% 5577 ± 4% brickland1/aim7/6000-page_test
> > 19542 ± 9% -38.3% 12057 ± 4% brickland1/aim7/6000-page_test
> > 993654 ± 2% -19.9% 795962 ± 3% brickland1/aim7/6000-page_test
>
> etc..
>
> how does one read that? afaict its a random number generator..
The "brickland1/aim7/6000-page_test" is the test case part.
The "TOTAL XXX" is the metric part. One test run may generate lots of
metrics, reflecting different aspect of the system dynamics.
This view may be easier to read, by grouping the metrics by test case.
test case: brickland1/aim7/6000-page_test
128529 ± 1% +17.9% 151594 ± 0% TOTAL aim7.jobs-per-min
582269 ±14% -55.6% 258617 ±16% TOTAL softirqs.SCHED
59366697 ± 3% -46.1% 32017187 ± 7% TOTAL cpuidle.C1-IVT.time
54543 ±11% -37.2% 34252 ±16% TOTAL cpuidle.C1-IVT.usage
2561 ± 7% -42.9% 1463 ± 9% TOTAL numa-numastat.node2.other_node
9926 ± 2% -43.8% 5577 ± 4% TOTAL proc-vmstat.numa_other
2627 ±12% -49.1% 1337 ±12% TOTAL numa-numastat.node1.other_node
19542 ± 9% -38.3% 12057 ± 4% TOTAL cpuidle.C1E-IVT.usage
2455 ±10% -41.0% 1448 ± 9% TOTAL numa-numastat.node0.other_node
471304 ±11% -31.4% 323251 ± 8% TOTAL numa-vmstat.node1.nr_anon_pages
2281 ±12% -41.8% 1327 ±16% TOTAL numa-numastat.node3.other_node
1903446 ±11% -30.7% 1318156 ± 7% TOTAL numa-meminfo.node1.AnonPages
518274 ±11% -30.4% 360742 ± 8% TOTAL numa-vmstat.node1.nr_active_anon
2097138 ±10% -30.0% 1469003 ± 8% TOTAL numa-meminfo.node1.Active(anon)
49527464 ± 6% -32.4% 33488833 ± 4% TOTAL cpuidle.C1E-IVT.time
2118206 ±10% -29.7% 1488874 ± 7% TOTAL numa-meminfo.node1.Active
76064 ± 3% -32.2% 51572 ± 6% TOTAL cpuidle.C6-IVT.usage
188938 ±33% -41.3% 110966 ±16% TOTAL numa-meminfo.node2.PageTables
47262 ±35% -42.3% 27273 ±16% TOTAL numa-vmstat.node2.nr_page_table_pages
1944687 ±10% -25.8% 1443923 ±16% TOTAL numa-meminfo.node3.Active(anon)
1754763 ±11% -26.6% 1288713 ±16% TOTAL numa-meminfo.node3.AnonPages
1964722 ±10% -25.5% 1464696 ±16% TOTAL numa-meminfo.node3.Active
432109 ± 9% -26.2% 318886 ±14% TOTAL numa-vmstat.node3.nr_anon_pages
479527 ± 9% -25.3% 358029 ±14% TOTAL numa-vmstat.node3.nr_active_anon
463719 ± 8% -24.7% 349388 ± 7% TOTAL numa-vmstat.node0.nr_anon_pages
3157742 ±16% -26.5% 2320253 ±10% TOTAL numa-meminfo.node1.MemUsed
7303589 ± 2% -24.8% 5495829 ± 3% TOTAL meminfo.AnonPages
8064024 ± 2% -24.0% 6132677 ± 3% TOTAL meminfo.Active(anon)
511455 ± 8% -23.9% 389447 ± 7% TOTAL numa-vmstat.node0.nr_active_anon
1818612 ± 2% -24.9% 1365670 ± 3% TOTAL proc-vmstat.nr_anon_pages
2007155 ± 2% -24.3% 1518688 ± 3% TOTAL proc-vmstat.nr_active_anon
8145316 ± 2% -23.7% 6213832 ± 3% TOTAL meminfo.Active
1850230 ± 8% -24.1% 1405061 ± 8% TOTAL numa-meminfo.node0.AnonPages
6.567e+11 ± 3% -21.4% 5.16e+11 ± 4% TOTAL meminfo.Committed_AS
2044097 ± 7% -23.5% 1562809 ± 8% TOTAL numa-meminfo.node0.Active(anon)
2064106 ± 7% -23.3% 1582792 ± 8% TOTAL numa-meminfo.node0.Active
235358 ± 5% -19.8% 188793 ± 3% TOTAL proc-vmstat.pgmigrate_success
235358 ± 5% -19.8% 188793 ± 3% TOTAL proc-vmstat.numa_pages_migrated
433235 ± 4% -18.1% 354845 ± 5% TOTAL numa-vmstat.node2.nr_anon_pages
198747 ±23% -28.0% 143034 ± 3% TOTAL proc-vmstat.nr_page_table_pages
3187 ± 5% -18.5% 2599 ± 6% TOTAL numa-vmstat.node0.numa_other
796281 ±23% -27.7% 575352 ± 3% TOTAL meminfo.PageTables
1395062 ± 6% -19.0% 1130108 ± 3% TOTAL proc-vmstat.numa_hint_faults
477037 ± 4% -17.2% 394983 ± 5% TOTAL numa-vmstat.node2.nr_active_anon
2829 ±10% +18.7% 3357 ± 3% TOTAL numa-vmstat.node2.nr_alloc_batch
993654 ± 2% -19.9% 795962 ± 3% TOTAL softirqs.RCU
2706 ± 4% +26.1% 3411 ± 5% TOTAL numa-vmstat.node1.nr_alloc_batch
2725835 ± 4% -17.5% 2247537 ± 4% TOTAL numa-meminfo.node2.MemUsed
393637 ± 6% -15.3% 333296 ± 2% TOTAL proc-vmstat.numa_hint_faults_local
2.82 ± 3% +21.9% 3.43 ± 4% TOTAL turbostat.%pc2
4.40 ± 2% +22.0% 5.37 ± 4% TOTAL turbostat.%c6
1742111 ± 4% -16.9% 1447181 ± 5% TOTAL numa-meminfo.node2.AnonPages
15865125 ± 1% -15.0% 13485882 ± 1% TOTAL softirqs.TIMER
1923000 ± 4% -16.4% 1608509 ± 5% TOTAL numa-meminfo.node2.Active(anon)
1943185 ± 4% -16.2% 1629057 ± 5% TOTAL numa-meminfo.node2.Active
3077 ± 1% +14.5% 3523 ± 0% TOTAL proc-vmstat.pgactivate
329 ± 1% -13.3% 285 ± 0% TOTAL uptime.boot
13158 ±13% -14.4% 11261 ± 4% TOTAL numa-meminfo.node3.SReclaimable
3289 ±13% -14.4% 2815 ± 4% TOTAL numa-vmstat.node3.nr_slab_reclaimable
3150464 ± 2% -24.2% 2387551 ± 3% TOTAL time.voluntary_context_switches
281 ± 1% -15.1% 238 ± 0% TOTAL time.elapsed_time
29294 ± 1% -14.3% 25093 ± 0% TOTAL time.system_time
4529818 ± 1% -8.8% 4129398 ± 1% TOTAL time.involuntary_context_switches
15.75 ± 1% -3.4% 15.21 ± 0% TOTAL turbostat.RAM_W
10655 ± 0% +1.4% 10802 ± 0% TOTAL time.percent_of_cpu_this_job_got
Thanks,
Fengguang
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [sched] 143e1e28cb4: +17.9% aim7.jobs-per-min, -9.7% hackbench.throughput
2014-08-10 10:54 ` Fengguang Wu
@ 2014-08-10 15:05 ` Peter Zijlstra
2014-08-10 15:16 ` Ingo Molnar
2014-08-11 1:23 ` Fengguang Wu
2014-08-11 13:33 ` Peter Zijlstra
1 sibling, 2 replies; 12+ messages in thread
From: Peter Zijlstra @ 2014-08-10 15:05 UTC (permalink / raw)
To: Fengguang Wu
Cc: Vincent Guittot, Dave Hansen, LKML, lkp, Ingo Molnar,
Dietmar Eggemann, Preeti U Murthy
[-- Attachment #1: Type: text/plain, Size: 758 bytes --]
On Sun, Aug 10, 2014 at 06:54:13PM +0800, Fengguang Wu wrote:
> The "brickland1/aim7/6000-page_test" is the test case part.
>
> The "TOTAL XXX" is the metric part. One test run may generate lots of
> metrics, reflecting different aspect of the system dynamics.
>
> This view may be easier to read, by grouping the metrics by test case.
Right, at least that makes more sense.
> test case: brickland1/aim7/6000-page_test
Ok, so next question, what is a brickland? I suspect its a machine of
sorts, seeing how some others had wsm in that part of the test. Now I
know what a westmere is, but I've never heard of a brickland.
Please describe the machine, this is a topology patch, so we need to
know the topology of the affected machines.
[-- Attachment #2: Type: application/pgp-signature, Size: 836 bytes --]
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [sched] 143e1e28cb4: +17.9% aim7.jobs-per-min, -9.7% hackbench.throughput
2014-08-10 15:05 ` Peter Zijlstra
@ 2014-08-10 15:16 ` Ingo Molnar
2014-08-11 1:23 ` Fengguang Wu
1 sibling, 0 replies; 12+ messages in thread
From: Ingo Molnar @ 2014-08-10 15:16 UTC (permalink / raw)
To: Peter Zijlstra
Cc: Fengguang Wu, Vincent Guittot, Dave Hansen, LKML, lkp,
Dietmar Eggemann, Preeti U Murthy
* Peter Zijlstra <peterz@infradead.org> wrote:
> On Sun, Aug 10, 2014 at 06:54:13PM +0800, Fengguang Wu wrote:
> > The "brickland1/aim7/6000-page_test" is the test case part.
> >
> > The "TOTAL XXX" is the metric part. One test run may generate lots of
> > metrics, reflecting different aspect of the system dynamics.
> >
> > This view may be easier to read, by grouping the metrics by test case.
>
> Right, at least that makes more sense.
>
> > test case: brickland1/aim7/6000-page_test
>
> Ok, so next question, what is a brickland? I suspect its a machine of
> sorts, seeing how some others had wsm in that part of the test. Now I
> know what a westmere is, but I've never heard of a brickland.
>
> Please describe the machine, this is a topology patch, so we need to
> know the topology of the affected machines.
Wikipedia says:
http://en.wikipedia.org/wiki/List_of_Intel_codenames
Brickland Platform High-end server platform based on the Ivy Bridge-EX processor.[6] Reference unknown. 2010
Thanks,
Ingo
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [sched] 143e1e28cb4: +17.9% aim7.jobs-per-min, -9.7% hackbench.throughput
2014-08-10 15:05 ` Peter Zijlstra
2014-08-10 15:16 ` Ingo Molnar
@ 2014-08-11 1:23 ` Fengguang Wu
2014-08-12 14:57 ` kodiak furr
1 sibling, 1 reply; 12+ messages in thread
From: Fengguang Wu @ 2014-08-11 1:23 UTC (permalink / raw)
To: Peter Zijlstra
Cc: Vincent Guittot, Dave Hansen, LKML, lkp, Ingo Molnar,
Dietmar Eggemann, Preeti U Murthy
On Sun, Aug 10, 2014 at 05:05:03PM +0200, Peter Zijlstra wrote:
> On Sun, Aug 10, 2014 at 06:54:13PM +0800, Fengguang Wu wrote:
> > The "brickland1/aim7/6000-page_test" is the test case part.
> >
> > The "TOTAL XXX" is the metric part. One test run may generate lots of
> > metrics, reflecting different aspect of the system dynamics.
> >
> > This view may be easier to read, by grouping the metrics by test case.
>
> Right, at least that makes more sense.
>
> > test case: brickland1/aim7/6000-page_test
>
> Ok, so next question, what is a brickland? I suspect its a machine of
> sorts, seeing how some others had wsm in that part of the test. Now I
> know what a westmere is, but I've never heard of a brickland.
As Ingo says, it's the "Brickland" platform with "Ivy Bridge-EX" CPU.
Sorry this part can be improved -- I'll add description of the test
boxes in future reports.
> Please describe the machine, this is a topology patch, so we need to
> know the topology of the affected machines.
Here they are. If you need more (or less) information (now and future),
please let me know.
brickland1: Brickland Ivy Bridge-EX
Memory: 128G
brickland3: Brickland Ivy Bridge-EX
Memory: 512G
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Byte Order: Little Endian
CPU(s): 120
On-line CPU(s) list: 0-119
Thread(s) per core: 2
Core(s) per socket: 15
Socket(s): 4
NUMA node(s): 4
Vendor ID: GenuineIntel
CPU family: 6
Model: 62
Stepping: 7
CPU MHz: 3192.875
BogoMIPS: 5593.49
Virtualization: VT-x
L1d cache: 32K
L1i cache: 32K
L2 cache: 256K
L3 cache: 38400K
NUMA node0 CPU(s): 0-14,60-74
NUMA node1 CPU(s): 15-29,75-89
NUMA node2 CPU(s): 30-44,90-104
NUMA node3 CPU(s): 45-59,105-119
lkp-snb01: Sandy Bridge-EP
Memory: 32G
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Byte Order: Little Endian
CPU(s): 32
On-line CPU(s) list: 0-31
Thread(s) per core: 2
Core(s) per socket: 8
Socket(s): 2
NUMA node(s): 2
Vendor ID: GenuineIntel
CPU family: 6
Model: 45
Stepping: 6
CPU MHz: 3498.820
BogoMIPS: 5391.31
Virtualization: VT-x
L1d cache: 32K
L1i cache: 32K
L2 cache: 256K
L3 cache: 20480K
NUMA node0 CPU(s): 0-7,16-23
NUMA node1 CPU(s): 8-15,24-31
wsm: Westmere
Memory: 6G
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Byte Order: Little Endian
CPU(s): 12
On-line CPU(s) list: 0-11
Thread(s) per core: 2
Core(s) per socket: 6
Socket(s): 1
NUMA node(s): 1
Vendor ID: GenuineIntel
CPU family: 6
Model: 44
Stepping: 2
CPU MHz: 3458.000
BogoMIPS: 6756.46
Virtualization: VT-x
L1d cache: 32K
L1i cache: 32K
L2 cache: 256K
L3 cache: 12288K
NUMA node0 CPU(s): 0-11
lkp-nex04: Nehalem-EX
Memory: 256G
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Byte Order: Little Endian
CPU(s): 64
On-line CPU(s) list: 0-63
Thread(s) per core: 2
Core(s) per socket: 8
Socket(s): 4
NUMA node(s): 4
Vendor ID: GenuineIntel
CPU family: 6
Model: 46
Stepping: 6
CPU MHz: 2262.000
BogoMIPS: 4521.30
Virtualization: VT-x
L1d cache: 32K
L1i cache: 32K
L2 cache: 256K
L3 cache: 24576K
NUMA node0 CPU(s): 0-7,32-39
NUMA node1 CPU(s): 8-15,40-47
NUMA node2 CPU(s): 16-23,48-55
NUMA node3 CPU(s): 24-31,56-63
nhm4: Nehalem
Memory: 4G
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Byte Order: Little Endian
CPU(s): 8
On-line CPU(s) list: 0-7
Thread(s) per core: 2
Core(s) per socket: 4
Socket(s): 1
NUMA node(s): 1
Vendor ID: GenuineIntel
CPU family: 6
Model: 26
Stepping: 4
CPU MHz: 3193.000
BogoMIPS: 6400.40
Virtualization: VT-x
L1d cache: 32K
L1i cache: 32K
L2 cache: 256K
L3 cache: 8192K
NUMA node0 CPU(s): 0-7
Thanks,
Fengguang
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [sched] 143e1e28cb4: +17.9% aim7.jobs-per-min, -9.7% hackbench.throughput
2014-08-10 10:54 ` Fengguang Wu
2014-08-10 15:05 ` Peter Zijlstra
@ 2014-08-11 13:33 ` Peter Zijlstra
2014-08-12 3:59 ` Preeti U Murthy
` (2 more replies)
1 sibling, 3 replies; 12+ messages in thread
From: Peter Zijlstra @ 2014-08-11 13:33 UTC (permalink / raw)
To: Fengguang Wu
Cc: Vincent Guittot, Dave Hansen, LKML, lkp, Ingo Molnar,
Dietmar Eggemann, Preeti U Murthy
[-- Attachment #1: Type: text/plain, Size: 4329 bytes --]
On Sun, Aug 10, 2014 at 06:54:13PM +0800, Fengguang Wu wrote:
> This view may be easier to read, by grouping the metrics by test case.
>
> test case: brickland1/aim7/6000-page_test
OK, I have a similar system to the brickland thing (slightly different
configuration, but should be close enough).
Now; do you have a description of each test-case someplace? In
particular, it might be good to have a small annotation to show which
direction is better.
>
> 128529 ± 1% +17.9% 151594 ± 0% TOTAL aim7.jobs-per-min
jobs per minute, + is better, so no worries there.
> 582269 ±14% -55.6% 258617 ±16% TOTAL softirqs.SCHED
> 993654 ± 2% -19.9% 795962 ± 3% TOTAL softirqs.RCU
> 15865125 ± 1% -15.0% 13485882 ± 1% TOTAL softirqs.TIMER
> 59366697 ± 3% -46.1% 32017187 ± 7% TOTAL cpuidle.C1-IVT.time
> 54543 ±11% -37.2% 34252 ±16% TOTAL cpuidle.C1-IVT.usage
> 19542 ± 9% -38.3% 12057 ± 4% TOTAL cpuidle.C1E-IVT.usage
> 49527464 ± 6% -32.4% 33488833 ± 4% TOTAL cpuidle.C1E-IVT.time
> 76064 ± 3% -32.2% 51572 ± 6% TOTAL cpuidle.C6-IVT.usage
Less idle time; might be good, if the work is cpubound, might be bad if
not; hard to say.
> 2.82 ± 3% +21.9% 3.43 ± 4% TOTAL turbostat.%pc2
> 4.40 ± 2% +22.0% 5.37 ± 4% TOTAL turbostat.%c6
> 15.75 ± 1% -3.4% 15.21 ± 0% TOTAL turbostat.RAM_W
> 3150464 ± 2% -24.2% 2387551 ± 3% TOTAL time.voluntary_context_switches
Typically less ctxsw is better..
> 281 ± 1% -15.1% 238 ± 0% TOTAL time.elapsed_time
> 29294 ± 1% -14.3% 25093 ± 0% TOTAL time.system_time
Less time spend (on presumably the same work) is better
> 4529818 ± 1% -8.8% 4129398 ± 1% TOTAL time.involuntary_context_switches
Less preemptions, also generally better
> 10655 ± 0% +1.4% 10802 ± 0% TOTAL time.percent_of_cpu_this_job_got
Seem an improvement; not sure.
Many more stats.. but from the above it looks like its an overall 'win';
or am I reading the thing wrong?
Now I think I see why this is; we've reduced load balancing frequency
significantly on this machine due to:
-#define SD_SIBLING_INIT (struct sched_domain) { \
- .min_interval = 1, \
- .max_interval = 2, \
-#define SD_MC_INIT (struct sched_domain) { \
- .min_interval = 1, \
- .max_interval = 4, \
-#define SD_CPU_INIT (struct sched_domain) { \
- .min_interval = 1, \
- .max_interval = 4, \
*sd = (struct sched_domain){
.min_interval = sd_weight,
.max_interval = 2*sd_weight,
Which both increased the min and max value significantly for all domains
involved.
That said; I think we might want to do something like the below; I can
imagine decreasing load balancing too much will negatively impact other
workloads.
Maybe slightly modified to make sure the first domain has a min_interval
of 1.
---
kernel/sched/core.c | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 1211575a2208..67ed5d854da1 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -6049,8 +6049,8 @@ sd_init(struct sched_domain_topology_level *tl, int cpu)
sd_flags &= ~TOPOLOGY_SD_FLAGS;
*sd = (struct sched_domain){
- .min_interval = sd_weight,
- .max_interval = 2*sd_weight,
+ .min_interval = max(1, sd_weight/2),
+ .max_interval = sd_weight,
.busy_factor = 32,
.imbalance_pct = 125,
@@ -6076,7 +6076,7 @@ sd_init(struct sched_domain_topology_level *tl, int cpu)
,
.last_balance = jiffies,
- .balance_interval = sd_weight,
+ .balance_interval = max(1, sd_weight/2),
.smt_gain = 0,
.max_newidle_lb_cost = 0,
.next_decay_max_lb_cost = jiffies,
[-- Attachment #2: Type: application/pgp-signature, Size: 836 bytes --]
^ permalink raw reply related [flat|nested] 12+ messages in thread
* Re: [sched] 143e1e28cb4: +17.9% aim7.jobs-per-min, -9.7% hackbench.throughput
2014-08-11 13:33 ` Peter Zijlstra
@ 2014-08-12 3:59 ` Preeti U Murthy
2014-08-12 6:41 ` Peter Zijlstra
2014-08-12 14:30 ` Fengguang Wu
2014-08-25 13:47 ` Vincent Guittot
2 siblings, 1 reply; 12+ messages in thread
From: Preeti U Murthy @ 2014-08-12 3:59 UTC (permalink / raw)
To: Peter Zijlstra, Fengguang Wu
Cc: Vincent Guittot, Dave Hansen, LKML, lkp, Ingo Molnar, Dietmar Eggemann
On 08/11/2014 07:03 PM, Peter Zijlstra wrote:
>
> Now I think I see why this is; we've reduced load balancing frequency
> significantly on this machine due to:
We have also changed the value of busy_factor to 32 from 64 across all
domains. This would contribute to increased frequency of load balancing?
Regards
Preeti U Murthy
>
>
> -#define SD_SIBLING_INIT (struct sched_domain) { \
> - .min_interval = 1, \
> - .max_interval = 2, \
>
>
> -#define SD_MC_INIT (struct sched_domain) { \
> - .min_interval = 1, \
> - .max_interval = 4, \
>
>
> -#define SD_CPU_INIT (struct sched_domain) { \
> - .min_interval = 1, \
> - .max_interval = 4, \
>
>
> *sd = (struct sched_domain){
> .min_interval = sd_weight,
> .max_interval = 2*sd_weight,
>
> Which both increased the min and max value significantly for all domains
> involved.
>
> That said; I think we might want to do something like the below; I can
> imagine decreasing load balancing too much will negatively impact other
> workloads.
>
> Maybe slightly modified to make sure the first domain has a min_interval
> of 1.
>
> ---
> kernel/sched/core.c | 6 +++---
> 1 file changed, 3 insertions(+), 3 deletions(-)
>
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index 1211575a2208..67ed5d854da1 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -6049,8 +6049,8 @@ sd_init(struct sched_domain_topology_level *tl, int cpu)
> sd_flags &= ~TOPOLOGY_SD_FLAGS;
>
> *sd = (struct sched_domain){
> - .min_interval = sd_weight,
> - .max_interval = 2*sd_weight,
> + .min_interval = max(1, sd_weight/2),
> + .max_interval = sd_weight,
> .busy_factor = 32,
> .imbalance_pct = 125,
>
> @@ -6076,7 +6076,7 @@ sd_init(struct sched_domain_topology_level *tl, int cpu)
> ,
>
> .last_balance = jiffies,
> - .balance_interval = sd_weight,
> + .balance_interval = max(1, sd_weight/2),
> .smt_gain = 0,
> .max_newidle_lb_cost = 0,
> .next_decay_max_lb_cost = jiffies,
>
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [sched] 143e1e28cb4: +17.9% aim7.jobs-per-min, -9.7% hackbench.throughput
2014-08-12 3:59 ` Preeti U Murthy
@ 2014-08-12 6:41 ` Peter Zijlstra
0 siblings, 0 replies; 12+ messages in thread
From: Peter Zijlstra @ 2014-08-12 6:41 UTC (permalink / raw)
To: Preeti U Murthy
Cc: Fengguang Wu, Vincent Guittot, Dave Hansen, LKML, lkp,
Ingo Molnar, Dietmar Eggemann
[-- Attachment #1: Type: text/plain, Size: 520 bytes --]
On Tue, Aug 12, 2014 at 09:29:19AM +0530, Preeti U Murthy wrote:
> On 08/11/2014 07:03 PM, Peter Zijlstra wrote:
> >
> > Now I think I see why this is; we've reduced load balancing frequency
> > significantly on this machine due to:
>
> We have also changed the value of busy_factor to 32 from 64 across all
> domains. This would contribute to increased frequency of load balancing?
Not enough to offset the changes in min/max. Lowering it makes sense
because we've increased min/max for most situations.
[-- Attachment #2: Type: application/pgp-signature, Size: 836 bytes --]
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [sched] 143e1e28cb4: +17.9% aim7.jobs-per-min, -9.7% hackbench.throughput
2014-08-11 13:33 ` Peter Zijlstra
2014-08-12 3:59 ` Preeti U Murthy
@ 2014-08-12 14:30 ` Fengguang Wu
2014-08-25 13:47 ` Vincent Guittot
2 siblings, 0 replies; 12+ messages in thread
From: Fengguang Wu @ 2014-08-12 14:30 UTC (permalink / raw)
To: Peter Zijlstra
Cc: Vincent Guittot, Dave Hansen, LKML, lkp, Ingo Molnar,
Dietmar Eggemann, Preeti U Murthy
On Mon, Aug 11, 2014 at 03:33:52PM +0200, Peter Zijlstra wrote:
> On Sun, Aug 10, 2014 at 06:54:13PM +0800, Fengguang Wu wrote:
> > This view may be easier to read, by grouping the metrics by test case.
> >
> > test case: brickland1/aim7/6000-page_test
>
> OK, I have a similar system to the brickland thing (slightly different
> configuration, but should be close enough).
>
> Now; do you have a description of each test-case someplace?
You can find our aim7 test script here:
git clone git://git.kernel.org/pub/scm/linux/kernel/git/wfg/lkp-tests
cd lkp-tests
vi tests/aim7
More test scripts are available there:
vi tests/hackbench
vi tests/netperf
...
> In particular, it might be good to have a small annotation to show
> which direction is better.
The directions are listed in these files as positive/negative numbers:
vi metric/index-*
For examples:
% head -3 metric/index-*
==> metric/index-latency.yaml <==
dbench.max_latency: -0.1
fileio.request_latency_95%_ms: -0.2
oltp.request_latency_95%_ms: -0.2
==> metric/index-perf.yaml <==
aim7.jobs-per-min: 1
dbench.throughput-MB/sec: 1
ebizzy.throughput: 1
==> metric/index-power.yaml <==
turbostat.Pkg_W: -1
turbostat.RAM_W: -1
turbostat.%c0: -0.1
==> metric/index-size.yaml <==
kernel-size.text: -1
kernel-size.data: -1
kernel-size.bss: -1
They are not the comprehensive list, but reasonably complete to list
the most important ones.
> > 128529 ± 1% +17.9% 151594 ± 0% TOTAL aim7.jobs-per-min
>
> jobs per minute, + is better, so no worries there.
>
> > 582269 ±14% -55.6% 258617 ±16% TOTAL softirqs.SCHED
> > 993654 ± 2% -19.9% 795962 ± 3% TOTAL softirqs.RCU
> > 15865125 ± 1% -15.0% 13485882 ± 1% TOTAL softirqs.TIMER
>
> > 59366697 ± 3% -46.1% 32017187 ± 7% TOTAL cpuidle.C1-IVT.time
> > 54543 ±11% -37.2% 34252 ±16% TOTAL cpuidle.C1-IVT.usage
> > 19542 ± 9% -38.3% 12057 ± 4% TOTAL cpuidle.C1E-IVT.usage
> > 49527464 ± 6% -32.4% 33488833 ± 4% TOTAL cpuidle.C1E-IVT.time
> > 76064 ± 3% -32.2% 51572 ± 6% TOTAL cpuidle.C6-IVT.usage
>
> Less idle time; might be good, if the work is cpubound, might be bad if
> not; hard to say.
>
> > 2.82 ± 3% +21.9% 3.43 ± 4% TOTAL turbostat.%pc2
> > 4.40 ± 2% +22.0% 5.37 ± 4% TOTAL turbostat.%c6
> > 15.75 ± 1% -3.4% 15.21 ± 0% TOTAL turbostat.RAM_W
>
> > 3150464 ± 2% -24.2% 2387551 ± 3% TOTAL time.voluntary_context_switches
>
> Typically less ctxsw is better..
>
> > 281 ± 1% -15.1% 238 ± 0% TOTAL time.elapsed_time
> > 29294 ± 1% -14.3% 25093 ± 0% TOTAL time.system_time
>
> Less time spend (on presumably the same work) is better
>
> > 4529818 ± 1% -8.8% 4129398 ± 1% TOTAL time.involuntary_context_switches
>
> Less preemptions, also generally better
>
> > 10655 ± 0% +1.4% 10802 ± 0% TOTAL time.percent_of_cpu_this_job_got
>
> Seem an improvement; not sure.
>
> Many more stats.. but from the above it looks like its an overall 'win';
> or am I reading the thing wrong?
I'd agree with your interpretations, too.
In case you want to make sure the exact meaning of the above values:
they are generated by scripts in stats/* and stats/hackbench would be
a good example to read.
Thanks,
Fengguang
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [sched] 143e1e28cb4: +17.9% aim7.jobs-per-min, -9.7% hackbench.throughput
2014-08-11 1:23 ` Fengguang Wu
@ 2014-08-12 14:57 ` kodiak furr
0 siblings, 0 replies; 12+ messages in thread
From: kodiak furr @ 2014-08-12 14:57 UTC (permalink / raw)
To: Fengguang Wu
Cc: lkp, Vincent Guittot, Dave Hansen, Peter Zijlstra,
Preeti U Murthy, LKML, Dietmar Eggemann, Ingo Molnar
[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset=utf-8, Size: 6665 bytes --]
.
Sent using Boxer
On Aug 10, 2014 8:23 PM, Fengguang Wu <fengguang.wu@intel.com> wrote:
>
> On Sun, Aug 10, 2014 at 05:05:03PM +0200, Peter Zijlstra wrote:
> > On Sun, Aug 10, 2014 at 06:54:13PM +0800, Fengguang Wu wrote:
> > > The "brickland1/aim7/6000-page_test" is the test case part.
> > >
> > > The "TOTAL XXX" is the metric part. One test run may generate lots of
> > > metrics, reflecting different aspect of the system dynamics.
> > >
> > > This view may be easier to read, by grouping the metrics by test case.
> >
> > Right, at least that makes more sense.
> >
> > > test case: brickland1/aim7/6000-page_test
> >
> > Ok, so next question, what is a brickland? I suspect its a machine of
> > sorts, seeing how some others had wsm in that part of the test. Now I
> > know what a westmere is, but I've never heard of a brickland.
>
> As Ingo says, it's the "Brickland" platform with "Ivy Bridge-EX" CPU.
>
> Sorry this part can be improved -- I'll add description of the test
> boxes in future reports.
>
> > Please describe the machine, this is a topology patch, so we need to
> > know the topology of the affected machines.
>
> Here they are. If you need more (or less) information (now and future),
> please let me know.
>
> brickland1: Brickland Ivy Bridge-EX
> Memory: 128G
>
> brickland3: Brickland Ivy Bridge-EX
> Memory: 512G
> Architecture:Â Â Â Â Â Â Â Â Â x86_64
> CPU op-mode(s):Â Â Â Â Â Â Â 32-bit, 64-bit
> Byte Order:Â Â Â Â Â Â Â Â Â Â Â Little Endian
> CPU(s):Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â 120
> On-line CPU(s) list:Â Â 0-119
> Thread(s) per core:Â Â Â 2
> Core(s) per socket:Â Â Â 15
> Socket(s):Â Â Â Â Â Â Â Â Â Â Â Â 4
> NUMA node(s):Â Â Â Â Â Â Â Â Â 4
> Vendor ID:Â Â Â Â Â Â Â Â Â Â Â Â GenuineIntel
> CPU family:Â Â Â Â Â Â Â Â Â Â Â 6
> Model:Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â 62
> Stepping:Â Â Â Â Â Â Â Â Â Â Â Â Â 7
> CPU MHz:Â Â Â Â Â Â Â Â Â Â Â Â Â Â 3192.875
> BogoMIPS:Â Â Â Â Â Â Â Â Â Â Â Â Â 5593.49
> Virtualization:Â Â Â Â Â Â Â VT-x
> L1d cache:Â Â Â Â Â Â Â Â Â Â Â Â 32K
> L1i cache:Â Â Â Â Â Â Â Â Â Â Â Â 32K
> L2 cache:Â Â Â Â Â Â Â Â Â Â Â Â Â 256K
> L3 cache:Â Â Â Â Â Â Â Â Â Â Â Â Â 38400K
> NUMA node0 CPU(s):Â Â Â Â 0-14,60-74
> NUMA node1 CPU(s):Â Â Â Â 15-29,75-89
> NUMA node2 CPU(s):Â Â Â Â 30-44,90-104
> NUMA node3 CPU(s):Â Â Â Â 45-59,105-119
>
> lkp-snb01: Sandy Bridge-EP
> Memory: 32G
> Architecture:Â Â Â Â Â Â Â Â Â x86_64
> CPU op-mode(s):Â Â Â Â Â Â Â 32-bit, 64-bit
> Byte Order:Â Â Â Â Â Â Â Â Â Â Â Little Endian
> CPU(s):Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â 32
> On-line CPU(s) list:Â Â 0-31
> Thread(s) per core:Â Â Â 2
> Core(s) per socket:Â Â Â 8
> Socket(s):Â Â Â Â Â Â Â Â Â Â Â Â 2
> NUMA node(s):Â Â Â Â Â Â Â Â Â 2
> Vendor ID:Â Â Â Â Â Â Â Â Â Â Â Â GenuineIntel
> CPU family:Â Â Â Â Â Â Â Â Â Â Â 6
> Model:Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â 45
> Stepping:Â Â Â Â Â Â Â Â Â Â Â Â Â 6
> CPU MHz:Â Â Â Â Â Â Â Â Â Â Â Â Â Â 3498.820
> BogoMIPS:Â Â Â Â Â Â Â Â Â Â Â Â Â 5391.31
> Virtualization:Â Â Â Â Â Â Â VT-x
> L1d cache:Â Â Â Â Â Â Â Â Â Â Â Â 32K
> L1i cache:Â Â Â Â Â Â Â Â Â Â Â Â 32K
> L2 cache:Â Â Â Â Â Â Â Â Â Â Â Â Â 256K
> L3 cache:Â Â Â Â Â Â Â Â Â Â Â Â Â 20480K
> NUMA node0 CPU(s):Â Â Â Â 0-7,16-23
> NUMA node1 CPU(s):Â Â Â Â 8-15,24-31
>
> wsm: Westmere
> Memory: 6G
> Architecture:Â Â Â Â Â Â Â Â Â x86_64
> CPU op-mode(s):Â Â Â Â Â Â Â 32-bit, 64-bit
> Byte Order:Â Â Â Â Â Â Â Â Â Â Â Little Endian
> CPU(s):Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â 12
> On-line CPU(s) list:Â Â 0-11
> Thread(s) per core:Â Â Â 2
> Core(s) per socket:Â Â Â 6
> Socket(s):Â Â Â Â Â Â Â Â Â Â Â Â 1
> NUMA node(s):Â Â Â Â Â Â Â Â Â 1
> Vendor ID:Â Â Â Â Â Â Â Â Â Â Â Â GenuineIntel
> CPU family:Â Â Â Â Â Â Â Â Â Â Â 6
> Model:Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â 44
> Stepping:Â Â Â Â Â Â Â Â Â Â Â Â Â 2
> CPU MHz:Â Â Â Â Â Â Â Â Â Â Â Â Â Â 3458.000
> BogoMIPS:Â Â Â Â Â Â Â Â Â Â Â Â Â 6756.46
> Virtualization:Â Â Â Â Â Â Â VT-x
> L1d cache:Â Â Â Â Â Â Â Â Â Â Â Â 32K
> L1i cache:Â Â Â Â Â Â Â Â Â Â Â Â 32K
> L2 cache:Â Â Â Â Â Â Â Â Â Â Â Â Â 256K
> L3 cache:Â Â Â Â Â Â Â Â Â Â Â Â Â 12288K
> NUMA node0 CPU(s):Â Â Â Â 0-11
>
> lkp-nex04: Nehalem-EX
> Memory: 256G
> Architecture:Â Â Â Â Â Â Â Â Â x86_64
> CPU op-mode(s):Â Â Â Â Â Â Â 32-bit, 64-bit
> Byte Order:Â Â Â Â Â Â Â Â Â Â Â Little Endian
> CPU(s):Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â 64
> On-line CPU(s) list:Â Â 0-63
> Thread(s) per core:Â Â Â 2
> Core(s) per socket:Â Â Â 8
> Socket(s):Â Â Â Â Â Â Â Â Â Â Â Â 4
> NUMA node(s):Â Â Â Â Â Â Â Â Â 4
> Vendor ID:Â Â Â Â Â Â Â Â Â Â Â Â GenuineIntel
> CPU family:Â Â Â Â Â Â Â Â Â Â Â 6
> Model:Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â 46
> Stepping:Â Â Â Â Â Â Â Â Â Â Â Â Â 6
> CPU MHz:Â Â Â Â Â Â Â Â Â Â Â Â Â Â 2262.000
> BogoMIPS:Â Â Â Â Â Â Â Â Â Â Â Â Â 4521.30
> Virtualization:Â Â Â Â Â Â Â VT-x
> L1d cache:Â Â Â Â Â Â Â Â Â Â Â Â 32K
> L1i cache:Â Â Â Â Â Â Â Â Â Â Â Â 32K
> L2 cache:Â Â Â Â Â Â Â Â Â Â Â Â Â 256K
> L3 cache:Â Â Â Â Â Â Â Â Â Â Â Â Â 24576K
> NUMA node0 CPU(s):Â Â Â Â 0-7,32-39
> NUMA node1 CPU(s):Â Â Â Â 8-15,40-47
> NUMA node2 CPU(s):Â Â Â Â 16-23,48-55
> NUMA node3 CPU(s):Â Â Â Â 24-31,56-63
>
> nhm4: Nehalem
> Memory: 4G
> Architecture:Â Â Â Â Â Â Â Â Â x86_64
> CPU op-mode(s):Â Â Â Â Â Â Â 32-bit, 64-bit
> Byte Order:Â Â Â Â Â Â Â Â Â Â Â Little Endian
> CPU(s):Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â 8
> On-line CPU(s) list:Â Â 0-7
> Thread(s) per core:Â Â Â 2
> Core(s) per socket:Â Â Â 4
> Socket(s):Â Â Â Â Â Â Â Â Â Â Â Â 1
> NUMA node(s):Â Â Â Â Â Â Â Â Â 1
> Vendor ID:Â Â Â Â Â Â Â Â Â Â Â Â GenuineIntel
> CPU family:Â Â Â Â Â Â Â Â Â Â Â 6
> Model:Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â 26
> Stepping:Â Â Â Â Â Â Â Â Â Â Â Â Â 4
> CPU MHz:Â Â Â Â Â Â Â Â Â Â Â Â Â Â 3193.000
> BogoMIPS:Â Â Â Â Â Â Â Â Â Â Â Â Â 6400.40
> Virtualization:Â Â Â Â Â Â Â VT-x
> L1d cache:Â Â Â Â Â Â Â Â Â Â Â Â 32K
> L1i cache:Â Â Â Â Â Â Â Â Â Â Â Â 32K
> L2 cache:Â Â Â Â Â Â Â Â Â Â Â Â Â 256K
> L3 cache:Â Â Â Â Â Â Â Â Â Â Â Â Â 8192K
> NUMA node0 CPU(s):Â Â Â Â 0-7
>
> Thanks,
> Fengguang
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
ÿôèº{.nÇ+‰·Ÿ®‰†+%ŠËÿ±éݶ\x17¥Šwÿº{.nÇ+‰·¥Š{±þG«éÿŠ{ayº\x1dʇڙë,j\a¢f£¢·hšïêÿ‘êçz_è®\x03(階ŠÝ¢j"ú\x1a¶^[m§ÿÿ¾\a«þG«éÿ¢¸?™¨èÚ&£ø§~á¶iO•æ¬z·švØ^\x14\x04\x1a¶^[m§ÿÿÃ\fÿ¶ìÿ¢¸?–I¥
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [sched] 143e1e28cb4: +17.9% aim7.jobs-per-min, -9.7% hackbench.throughput
2014-08-11 13:33 ` Peter Zijlstra
2014-08-12 3:59 ` Preeti U Murthy
2014-08-12 14:30 ` Fengguang Wu
@ 2014-08-25 13:47 ` Vincent Guittot
2 siblings, 0 replies; 12+ messages in thread
From: Vincent Guittot @ 2014-08-25 13:47 UTC (permalink / raw)
To: Peter Zijlstra, Fengguang Wu
Cc: Dave Hansen, LKML, lkp, Ingo Molnar, Dietmar Eggemann, Preeti U Murthy
On 11 August 2014 15:33, Peter Zijlstra <peterz@infradead.org> wrote:
[snip]
>
>
> Now I think I see why this is; we've reduced load balancing frequency
> significantly on this machine due to:
I agree with you about the root cause
>
>
> -#define SD_SIBLING_INIT (struct sched_domain) { \
> - .min_interval = 1, \
> - .max_interval = 2, \
>
>
> -#define SD_MC_INIT (struct sched_domain) { \
> - .min_interval = 1, \
> - .max_interval = 4, \
>
>
> -#define SD_CPU_INIT (struct sched_domain) { \
> - .min_interval = 1, \
> - .max_interval = 4, \
>
>
> *sd = (struct sched_domain){
> .min_interval = sd_weight,
> .max_interval = 2*sd_weight,
>
> Which both increased the min and max value significantly for all domains
> involved.
>
> That said; I think we might want to do something like the below; I can
> imagine decreasing load balancing too much will negatively impact other
> workloads.
>
> Maybe slightly modified to make sure the first domain has a min_interval
> of 1.
>
> ---
> kernel/sched/core.c | 6 +++---
> 1 file changed, 3 insertions(+), 3 deletions(-)
>
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index 1211575a2208..67ed5d854da1 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -6049,8 +6049,8 @@ sd_init(struct sched_domain_topology_level *tl, int cpu)
> sd_flags &= ~TOPOLOGY_SD_FLAGS;
>
> *sd = (struct sched_domain){
> - .min_interval = sd_weight,
> - .max_interval = 2*sd_weight,
> + .min_interval = max(1, sd_weight/2),
> + .max_interval = sd_weight,
What about using ilog2 like the default scaling mode of other
scheduler tunables ?
Regards,
Vincent
> .busy_factor = 32,
> .imbalance_pct = 125,
>
> @@ -6076,7 +6076,7 @@ sd_init(struct sched_domain_topology_level *tl, int cpu)
> ,
>
> .last_balance = jiffies,
> - .balance_interval = sd_weight,
> + .balance_interval = max(1, sd_weight/2),
> .smt_gain = 0,
> .max_newidle_lb_cost = 0,
> .next_decay_max_lb_cost = jiffies,
^ permalink raw reply [flat|nested] 12+ messages in thread
end of thread, other threads:[~2014-08-25 13:48 UTC | newest]
Thread overview: 12+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2014-08-10 4:41 [sched] 143e1e28cb4: +17.9% aim7.jobs-per-min, -9.7% hackbench.throughput Fengguang Wu
2014-08-10 7:59 ` Peter Zijlstra
2014-08-10 10:54 ` Fengguang Wu
2014-08-10 15:05 ` Peter Zijlstra
2014-08-10 15:16 ` Ingo Molnar
2014-08-11 1:23 ` Fengguang Wu
2014-08-12 14:57 ` kodiak furr
2014-08-11 13:33 ` Peter Zijlstra
2014-08-12 3:59 ` Preeti U Murthy
2014-08-12 6:41 ` Peter Zijlstra
2014-08-12 14:30 ` Fengguang Wu
2014-08-25 13:47 ` Vincent Guittot
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).