[LKP] [net] 11ef7a8996d: +27.3% netperf.Throughput_Mbps

From: Aaron Lu
Date: Fri Jul 25 2014 - 02:31:25 EST


FYI, we noticed the below changes on

commit 11ef7a8996d5d433c9cd75d80651297eccbf6d42 ("net: Performance fix for process_backlog")

test case: lkp-t410/netperf/300s-200%-10K-SCTP_STREAM_MANY

68b7107b62983f2 11ef7a8996d5d433c9cd75d80
--------------- -------------------------
1023 ~ 3% +27.3% 1302 ~ 0% TOTAL netperf.Throughput_Mbps
0.72 ~10% -91.4% 0.06 ~23% TOTAL turbostat.%c3
13385 ~12% -92.6% 987 ~12% TOTAL cpuidle.C6-NHM.usage
22745 ~10% -95.5% 1016 ~40% TOTAL cpuidle.C3-NHM.usage
42675736 ~12% -84.6% 6571705 ~19% TOTAL cpuidle.C6-NHM.time
88342 ~11% -82.1% 15811 ~ 9% TOTAL softirqs.SCHED
19148006 ~12% -95.7% 821873 ~27% TOTAL cpuidle.C3-NHM.time
439.94 ~10% -77.5% 99.05 ~ 5% TOTAL uptime.idle
1.35 ~ 6% -65.8% 0.46 ~24% TOTAL turbostat.%c6
4 ~23% +114.3% 9 ~ 0% TOTAL vmstat.procs.r
1680 ~ 3% +40.4% 2359 ~ 3% TOTAL proc-vmstat.nr_alloc_batch
447921 ~ 4% +36.2% 610047 ~ 0% TOTAL softirqs.TIMER
9.09 ~ 9% +27.6% 11.60 ~ 8% TOTAL perf-profile.cpu-cycles.memcpy.sctp_outq_flush.sctp_outq_uncork.sctp_cmd_interpreter.sctp_do_sm
2350916 ~ 4% +35.4% 3184088 ~ 0% TOTAL proc-vmstat.pgalloc_dma
7.82 ~ 9% +24.1% 9.71 ~ 4% TOTAL perf-profile.cpu-cycles.copy_user_generic_string.sctp_user_addto_chunk.sctp_datamsg_from_user.sctp_sendmsg.inet_sendmsg
38837537 ~ 3% +27.1% 49358639 ~ 0% TOTAL proc-vmstat.numa_local
38837537 ~ 3% +27.1% 49358639 ~ 0% TOTAL proc-vmstat.numa_hit
50216 ~ 1% +17.0% 58745 ~ 6% TOTAL softirqs.RCU
1.41 ~ 4% +11.6% 1.58 ~ 1% TOTAL perf-profile.cpu-cycles.get_page_from_freelist.__alloc_pages_nodemask.alloc_kmem_pages_node.kmalloc_large_node.__kmalloc_node_track_caller
48539 ~ 1% -25.5% 36171 ~ 1% TOTAL vmstat.system.cs
75.64 ~ 4% +30.6% 98.82 ~ 0% TOTAL turbostat.%c0
3949 ~ 2% +13.4% 4478 ~ 0% TOTAL vmstat.system.in

Legend:
~XX% - stddev percent
[+-]XX% - change percent


vmstat.system.cs

52000 ++------------------------------------------------------------------+
50000 ++ * .* *. ** *. * *. |
| .* + *.** *. : **.* .* .**.**. * + + *.* * + : * .**. *. .**
48000 ** * * * * * * * .* * * * ** |
46000 ++ * |
| |
44000 ++ |
42000 ++ |
40000 ++ |
| |
38000 +O O O |
36000 O+ O O O O OO OO OO O O O O O OO O OO O |
| O O O O O O O |
34000 ++ O O O O |
32000 ++------------------------------------------------------------------+


[*] bisect-good sample
[O] bisect-bad sample


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,
Aaron
echo performance > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
echo performance > /sys/devices/system/cpu/cpu1/cpufreq/scaling_governor
echo performance > /sys/devices/system/cpu/cpu2/cpufreq/scaling_governor
echo performance > /sys/devices/system/cpu/cpu3/cpufreq/scaling_governor
netserver
netperf -t SCTP_STREAM_MANY -c -C -l 300 -- -m 10K
netperf -t SCTP_STREAM_MANY -c -C -l 300 -- -m 10K
netperf -t SCTP_STREAM_MANY -c -C -l 300 -- -m 10K
netperf -t SCTP_STREAM_MANY -c -C -l 300 -- -m 10K
netperf -t SCTP_STREAM_MANY -c -C -l 300 -- -m 10K
netperf -t SCTP_STREAM_MANY -c -C -l 300 -- -m 10K
netperf -t SCTP_STREAM_MANY -c -C -l 300 -- -m 10K
netperf -t SCTP_STREAM_MANY -c -C -l 300 -- -m 10K