Messages in this thread | | | Date | Thu, 16 Jan 2014 09:35:08 +0800 | From | Fengguang Wu <> | Subject | [sched] b75e43bb77d: +87% hackbench.throughput |
| |
Alex,
We noticed that commit b75e43bb77d ("sched: find the latest idle cpu") increased hackbench throughput by +87% in test case hackbench 1600%-threads-pipe on a 2S SNB server.
v3.13-rc7 b75e43bb77df14e2209532c1e --------------- ------------------------- 171792 ~ 0% +87.0% 321176 ~ 0% hackbench.throughput 4677274 ~29% +706.1% 37705344 ~ 1% proc-vmstat.pgalloc_dma32 12718244 ~34% +700.6% 1.018e+08 ~ 3% numa-vmstat.node0.numa_local 12741758 ~34% +699.3% 1.019e+08 ~ 3% numa-vmstat.node0.numa_hit 275020 ~16% +535.4% 1747494 ~ 3% proc-vmstat.nr_tlb_remote_flush_received 24343551 ~30% +745.7% 2.059e+08 ~ 1% numa-numastat.node0.local_node 578758 ~13% +507.2% 3514489 ~ 2% interrupts.TLB 8.696e+08 ~ 1% -85.7% 1.246e+08 ~11% interrupts.RES 24343552 ~30% +745.7% 2.059e+08 ~ 1% numa-numastat.node0.numa_hit 24275 ~12% +296.2% 96175 ~ 4% proc-vmstat.nr_tlb_remote_flush 16816 ~ 1% -60.8% 6596 ~ 3% proc-vmstat.nr_tlb_local_flush_all 1.807e+08 ~ 1% +135.8% 4.26e+08 ~ 1% proc-vmstat.numa_local 1.807e+08 ~ 1% +135.8% 4.26e+08 ~ 1% proc-vmstat.numa_hit 1.818e+08 ~ 1% +135.4% 4.28e+08 ~ 1% proc-vmstat.pgfree 1.771e+08 ~ 1% +120.3% 3.903e+08 ~ 1% proc-vmstat.pgalloc_normal 270 ~14% +149.1% 673 ~ 5% slabinfo.mnt_cache.active_objs 270 ~14% +149.1% 673 ~ 5% slabinfo.mnt_cache.num_objs 378 ~ 2% -45.5% 206 ~ 5% proc-vmstat.nr_dirtied 1284432 ~ 2% -42.0% 744640 ~ 2% softirqs.RCU 4821300 ~ 1% -40.6% 2863710 ~ 1% proc-vmstat.pgfault 1.564e+08 ~ 4% +40.8% 2.202e+08 ~ 1% numa-numastat.node1.local_node 1.564e+08 ~ 4% +40.8% 2.202e+08 ~ 1% numa-numastat.node1.numa_hit 78523323 ~ 5% +39.2% 1.093e+08 ~ 3% numa-vmstat.node1.numa_local 78615652 ~ 5% +39.2% 1.094e+08 ~ 3% numa-vmstat.node1.numa_hit 7878 ~ 0% +16.4% 9168 ~ 1% vmstat.procs.r 1413677 ~ 0% -84.6% 218164 ~10% vmstat.system.in 2.386e+09 ~ 0% -76.1% 5.709e+08 ~ 6% time.voluntary_context_switches 5575434 ~ 0% -75.6% 1359735 ~ 5% vmstat.system.cs 9.359e+08 ~ 1% -73.2% 2.508e+08 ~ 6% time.involuntary_context_switches 1229364 ~ 1% +76.9% 2174598 ~ 0% time.minor_page_faults 1638 ~ 1% +60.0% 2622 ~ 1% time.user_time 17188 ~ 0% -8.4% 15749 ~ 0% time.system_time
hackbench.throughput
340000 ++----------------------------------------------------------------+ | O O O O O O O O O O | 320000 ++ O O | 300000 ++ | | | 280000 ++ | 260000 O+ O O O O | | | 240000 ++ | 220000 ++ | | | 200000 ++ | 180000 ++ | *...*...*...*...*..*...*...*...*...*...*...*...*..*...*...*...*...* 160000 ++----------------------------------------------------------------+
time.involuntary_context_switches
1e+09 ++-----------------------------------------------------------------+ *...*...*...*... ..*...*...*..*... ..*... | 9e+08 ++ *...*. *...*...*...*...*. *...* 8e+08 ++ | | | 7e+08 ++ | | | 6e+08 ++ | | | 5e+08 ++ | 4e+08 ++ | | | 3e+08 ++ | O O O O O O O O O O O O O O O O O | 2e+08 ++-----------------------------------------------------------------+
Thanks, Fengguang
| |