Thanks, Fengguang. As what I understand, this is pretty good result, right? Only I did not expect to be so big an impact. Yuyang -----Original Message----- From: Wu, Fengguang Sent: Thursday, June 26, 2014 10:06 AM To: Du, Yuyang Cc: Hansen, Dave; lkp(a)01.org; Chen, Jet Subject: [sched] cec017ed87c: -35.9% fileio.requests_per_sec Yuyang, FYI, we noticed the below changes on git://bee.sh.intel.com/git/ydu19/linux cc-v3-on-3.15-rc8 commit cec017ed87c7e5298fabaf1718f4f3c0fada0adb ("Remove update_rq_runnable_avg") Test case: snb-drag/fileio/600s-100%-1HDD-xfs-64G-1024f-seqrewr-sync v3.15-rc8 cec017ed87c7e5298fabaf171 --------------- ------------------------- 6522 ~ 0% -35.9% 4183 ~ 0% TOTAL fileio.requests_per_sec 26744 ~11% -87.8% 3251 ~ 0% TOTAL proc-vmstat.workingset_nodereclaim 117412 ~ 0% -74.7% 29747 ~ 0% TOTAL proc-vmstat.slabs_scanned 1833 ~ 3% -63.8% 664 ~ 5% TOTAL proc-vmstat.kswapd_high_wmark_hit_quickly 2959 ~ 2% -45.6% 1610 ~ 2% TOTAL proc-vmstat.pageoutrun 7158189 ~ 0% -39.1% 4359933 ~ 0% TOTAL proc-vmstat.pgsteal_kswapd_dma32 7225747 ~ 0% -39.0% 4410503 ~ 0% TOTAL proc-vmstat.pgsteal_kswapd_normal 7202579 ~ 0% -38.8% 4406846 ~ 0% TOTAL proc-vmstat.pgscan_kswapd_dma32 7272151 ~ 0% -38.7% 4461197 ~ 0% TOTAL proc-vmstat.pgscan_kswapd_normal 3305425 ~ 0% -37.9% 2052786 ~ 0% TOTAL cpuidle.C1-SNB.usage 15453350 ~ 0% -36.5% 9816203 ~ 0% TOTAL proc-vmstat.nr_written 15454708 ~ 0% -36.6% 9790561 ~ 0% TOTAL proc-vmstat.pgfree 61898955 ~ 0% -36.4% 39352814 ~ 0% TOTAL proc-vmstat.pgpgout 62643 ~ 0% -35.9% 40129 ~ 0% TOTAL softirqs.BLOCK 15654950 ~ 0% -35.9% 10040787 ~ 0% TOTAL proc-vmstat.nr_dirtied 16729217 ~ 0% -33.9% 11065396 ~ 0% TOTAL proc-vmstat.numa_hit 16729217 ~ 0% -33.9% 11065396 ~ 0% TOTAL proc-vmstat.numa_local 8354814 ~ 0% -33.8% 5534438 ~ 0% TOTAL proc-vmstat.pgalloc_dma32 8419660 ~ 0% -33.8% 5575636 ~ 0% TOTAL proc-vmstat.pgalloc_normal 42342300 ~ 3% -31.0% 29195355 ~ 3% TOTAL cpuidle.C3-SNB.time 40089990 ~ 3% -28.1% 28834724 ~ 2% TOTAL cpuidle.C1E-SNB.time 26312 ~ 3% -27.6% 19054 ~ 2% TOTAL cpuidle.C3-SNB.usage 38112 ~ 3% -19.3% 30755 ~ 1% TOTAL cpuidle.C1E-SNB.usage 1122 ~ 1% +11.2% 1248 ~ 3% TOTAL proc-vmstat.pgactivate 330418 ~ 4% -11.7% 291792 ~ 1% TOTAL cpuidle.C6-SNB.usage 650 ~11% +69.8% 1103 ~ 0% TOTAL iostat.sda.w_await 650 ~11% +69.8% 1103 ~ 0% TOTAL iostat.sda.await 3540598 ~ 0% -35.9% 2269116 ~ 0% TOTAL time.voluntary_context_switches 1.252e+08 ~ 0% -35.9% 80326643 ~ 0% TOTAL time.file_system_outputs 94546 ~11% -31.3% 64935 ~ 0% TOTAL iostat.sda.wkB/s 186 ~11% -31.3% 128 ~ 0% TOTAL iostat.sda.w/s 94678 ~11% -31.3% 65072 ~ 0% TOTAL vmstat.io.bo 50.09 ~ 0% -34.6% 32.75 ~ 0% TOTAL time.system_time 11273 ~11% -29.4% 7964 ~ 0% TOTAL vmstat.system.cs 1.56 ~11% -27.8% 1.12 ~ 0% TOTAL iostat.sda.wrqm/s 131.77 ~11% +8.1% 142.40 ~ 0% TOTAL iostat.sda.avgqu-sz 662 ~14% -9.0% 602 ~ 0% TOTAL time.elapsed_time Legend: ~XX% - stddev percent [+-]XX% - change percent time.system_time 55 ++---------------------------------------------------------------------+ | | 50 *+...*...*....*....*....*...* | 45 ++ | | | 40 ++ | | | 35 ++ | | O O O O O O O O O O O 30 ++ | 25 ++ | | | 20 ++ | O O O O O | 15 ++---------------------------------------------------------------------+ time.voluntary_context_switches 4e+06 ++----------------------------------------------------------------+ | | 3.5e+06 *+..*....*...*....*...*...* | 3e+06 ++ | | | 2.5e+06 ++ | | O O O O O O O O O O O 2e+06 ++ | | | 1.5e+06 ++ | 1e+06 ++ | | | 500000 ++ | O O O O O | 0 ++----------------------------------------------------------------+ time.file_system_outputs 1.3e+08 ++---------------------------------------------------------------+ 1.25e+08 *+..*....*...*...*....*...* | | | 1.2e+08 ++ | 1.15e+08 ++ | | | 1.1e+08 ++ | 1.05e+08 ++ | 1e+08 ++ | | | 9.5e+07 ++ | 9e+07 ++ | | | 8.5e+07 ++ | 8e+07 O+--O----O---O---O----O---O---O----O---O---O----O---O---O----O---O fileio.requests_per_sec 7000 ++-------------------------------------------------------------------+ | | 6500 *+...*...*....*...*....*....* | | | | | 6000 ++ | | | 5500 ++ | | | 5000 ++ | | | | | 4500 ++ | O O O O O O O O O O O O O O O O 4000 ++-------------------------------------------------------------------+ iostat.sda.await 1200 ++-------------------------------------------------------------------+ | | 1100 O+ O O O O O O O O O O O O O O O 1000 ++ | | | 900 ++ | | | 800 ++ | | | 700 ++ ..*...*....*...*. * | 600 *+. .. . | | . .. | 500 ++ . | | * | 400 ++-------------------------------------------------------------------+ iostat.sda.w_await 1200 ++-------------------------------------------------------------------+ | | 1100 O+ O O O O O O O O O O O O O O O 1000 ++ | | | 900 ++ | | | 800 ++ | | | 700 ++ ..*...*....*...*. * | 600 *+. .. . | | . .. | 500 ++ . | | * | 400 ++-------------------------------------------------------------------+ [*] 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, Fengguang

Then, that is interesting and confusing... By removing some useless codes (in critical patch) in scheduler, a regression? Yuyang -----Original Message----- From: Wu, Fengguang Sent: Thursday, June 26, 2014 10:26 AM To: Du, Yuyang Cc: Hansen, Dave; lkp(a)01.org; Chen, Jet Subject: Re: [sched] cec017ed87c: -35.9% fileio.requests_per_sec Yuyang, On Thu, Jun 26, 2014 at 10:20:14AM +0800, Du, Yuyang wrote: Unfortunately it looks like a performance regression: the fileio throughput dropped by 35.9%. Thanks, Fengguang

On Thu, Jun 26, 2014 at 10:35:44AM +0800, Du, Yuyang wrote: It's comparison for the single patch. Thanks, Fengguang

On Thu, Jun 26, 2014 at 12:33:41PM +0800, Du, Yuyang wrote: Simple retest will not help, as you may see, the system already tested it several times and the stddev is 0%. However it may show different results if rebase the patches to latest RC kernel and retest. Thanks, Fengguang > -----Original Message----- > From: Wu, Fengguang > Sent: Thursday, June 26, 2014 12:11 PM > To: Du, Yuyang > Cc: Hansen, Dave; lkp(a)01.org; Chen, Jet > Subject: Re: [sched] cec017ed87c: -35.9% fileio.requests_per_sec > > On Thu, Jun 26, 2014 at 10:35:44AM +0800, Du, Yuyang wrote: > > Oh, is it for the CC patchset or just "Remove update_rq_runnable_avg"? > > It's comparison for the single patch. > > Thanks, > Fengguang > > > > -----Original Message----- > > From: Wu, Fengguang > > Sent: Thursday, June 26, 2014 10:26 AM > > To: Du, Yuyang > > Cc: Hansen, Dave; lkp(a)01.org; Chen, Jet > > Subject: Re: [sched] cec017ed87c: -35.9% fileio.requests_per_sec > > > > Yuyang, > > > > On Thu, Jun 26, 2014 at 10:20:14AM +0800, Du, Yuyang wrote: > > > Thanks, Fengguang. As what I understand, this is pretty good result, right? Only I did not expect to be so big an impact. > > > > Unfortunately it looks like a performance regression: the fileio throughput dropped by 35.9%. > > > > Thanks, > > Fengguang > > > > > -----Original Message----- > > > From: Wu, Fengguang > > > Sent: Thursday, June 26, 2014 10:06 AM > > > To: Du, Yuyang > > > Cc: Hansen, Dave; lkp(a)01.org; Chen, Jet > > > Subject: [sched] cec017ed87c: -35.9% fileio.requests_per_sec > > > > > > Yuyang, > > > > > > FYI, we noticed the below changes on > > > > > > git://bee.sh.intel.com/git/ydu19/linux cc-v3-on-3.15-rc8 commit > > > cec017ed87c7e5298fabaf1718f4f3c0fada0adb ("Remove > > > update_rq_runnable_avg") > > > > > > Test case: snb-drag/fileio/600s-100%-1HDD-xfs-64G-1024f-seqrewr-sync > > > > > > v3.15-rc8 cec017ed87c7e5298fabaf171 > > > --------------- ------------------------- > > > > 26744 ~11% -87.8% 3251 ~ 0% TOTAL proc-vmstat.workingset_nodereclaim > > > 117412 ~ 0% -74.7% 29747 ~ 0% TOTAL proc-vmstat.slabs_scanned > > > 1833 ~ 3% -63.8% 664 ~ 5% TOTAL proc-vmstat.kswapd_high_wmark_hit_quickly > > > 2959 ~ 2% -45.6% 1610 ~ 2% TOTAL proc-vmstat.pageoutrun > > > 7158189 ~ 0% -39.1% 4359933 ~ 0% TOTAL proc-vmstat.pgsteal_kswapd_dma32 > > > 7225747 ~ 0% -39.0% 4410503 ~ 0% TOTAL proc-vmstat.pgsteal_kswapd_normal > > > 7202579 ~ 0% -38.8% 4406846 ~ 0% TOTAL proc-vmstat.pgscan_kswapd_dma32 > > > 7272151 ~ 0% -38.7% 4461197 ~ 0% TOTAL proc-vmstat.pgscan_kswapd_normal > > > 3305425 ~ 0% -37.9% 2052786 ~ 0% TOTAL cpuidle.C1-SNB.usage > > > 15453350 ~ 0% -36.5% 9816203 ~ 0% TOTAL proc-vmstat.nr_written > > > 15454708 ~ 0% -36.6% 9790561 ~ 0% TOTAL proc-vmstat.pgfree > > > 61898955 ~ 0% -36.4% 39352814 ~ 0% TOTAL proc-vmstat.pgpgout > > > 62643 ~ 0% -35.9% 40129 ~ 0% TOTAL softirqs.BLOCK > > > 15654950 ~ 0% -35.9% 10040787 ~ 0% TOTAL proc-vmstat.nr_dirtied > > > 16729217 ~ 0% -33.9% 11065396 ~ 0% TOTAL proc-vmstat.numa_hit > > > 16729217 ~ 0% -33.9% 11065396 ~ 0% TOTAL proc-vmstat.numa_local > > > 8354814 ~ 0% -33.8% 5534438 ~ 0% TOTAL proc-vmstat.pgalloc_dma32 > > > 8419660 ~ 0% -33.8% 5575636 ~ 0% TOTAL proc-vmstat.pgalloc_normal > > > 42342300 ~ 3% -31.0% 29195355 ~ 3% TOTAL cpuidle.C3-SNB.time > > > 40089990 ~ 3% -28.1% 28834724 ~ 2% TOTAL cpuidle.C1E-SNB.time > > > 26312 ~ 3% -27.6% 19054 ~ 2% TOTAL cpuidle.C3-SNB.usage > > > 38112 ~ 3% -19.3% 30755 ~ 1% TOTAL cpuidle.C1E-SNB.usage > > > 1122 ~ 1% +11.2% 1248 ~ 3% TOTAL proc-vmstat.pgactivate > > > 330418 ~ 4% -11.7% 291792 ~ 1% TOTAL cpuidle.C6-SNB.usage > > > 650 ~11% +69.8% 1103 ~ 0% TOTAL iostat.sda.w_await > > > 650 ~11% +69.8% 1103 ~ 0% TOTAL iostat.sda.await > > > 3540598 ~ 0% -35.9% 2269116 ~ 0% TOTAL time.voluntary_context_switches > > > 1.252e+08 ~ 0% -35.9% 80326643 ~ 0% TOTAL time.file_system_outputs > > > 94546 ~11% -31.3% 64935 ~ 0% TOTAL iostat.sda.wkB/s > > > 186 ~11% -31.3% 128 ~ 0% TOTAL iostat.sda.w/s > > > 94678 ~11% -31.3% 65072 ~ 0% TOTAL vmstat.io.bo > > > 50.09 ~ 0% -34.6% 32.75 ~ 0% TOTAL time.system_time > > > 11273 ~11% -29.4% 7964 ~ 0% TOTAL vmstat.system.cs > > > 1.56 ~11% -27.8% 1.12 ~ 0% TOTAL iostat.sda.wrqm/s > > > 131.77 ~11% +8.1% 142.40 ~ 0% TOTAL iostat.sda.avgqu-sz > > > 662 ~14% -9.0% 602 ~ 0% TOTAL time.elapsed_time > > > > > > Legend: > > > ~XX% - stddev percent > > > [+-]XX% - change percent > > > > > > > > > time.system_time > > > > > > 55 ++---------------------------------------------------------------------+ > > > | | > > > 50 *+...*...*....*....*....*...* | > > > 45 ++ | > > > | | > > > 40 ++ | > > > | | > > > 35 ++ | > > > | O O O O O O O O O O O > > > 30 ++ | > > > 25 ++ | > > > | | > > > 20 ++ | > > > O O O O O | > > > 15 > > > ++------------------------------------------------------------------ > > > ++-- > > > -+ > > > > > > > > > time.voluntary_context_switches > > > > > > 4e+06 ++----------------------------------------------------------------+ > > > | | > > > 3.5e+06 *+..*....*...*....*...*...* | > > > 3e+06 ++ | > > > | | > > > 2.5e+06 ++ | > > > | O O O O O O O O O O O > > > 2e+06 ++ | > > > | | > > > 1.5e+06 ++ | > > > 1e+06 ++ | > > > | | > > > 500000 ++ | > > > O O O O O | > > > 0 > > > ++----------------------------------------------------------------+ > > > > > > > > > time.file_system_outputs > > > > > > 1.3e+08 ++---------------------------------------------------------------+ > > > 1.25e+08 *+..*....*...*...*....*...* | > > > | | > > > 1.2e+08 ++ | > > > 1.15e+08 ++ | > > > | | > > > 1.1e+08 ++ | > > > 1.05e+08 ++ | > > > 1e+08 ++ | > > > | | > > > 9.5e+07 ++ | > > > 9e+07 ++ | > > > | | > > > 8.5e+07 ++ | > > > 8e+07 > > > O+--O----O---O---O----O---O---O----O---O---O----O---O---O----O---O > > > > > > > > > fileio.requests_per_sec > > > > > > 7000 ++-------------------------------------------------------------------+ > > > | | > > > 6500 *+...*...*....*...*....*....* | > > > | | > > > | | > > > 6000 ++ | > > > | | > > > 5500 ++ | > > > | | > > > 5000 ++ | > > > | | > > > | | > > > 4500 ++ | > > > O O O O O O O O O O O O O O O O > > > 4000 > > > ++-------------------------------------------------------------------+ > > > > > > > > > iostat.sda.await > > > > > > 1200 ++-------------------------------------------------------------------+ > > > | | > > > 1100 O+ O O O O O O O O O O O O O O O > > > 1000 ++ | > > > | | > > > 900 ++ | > > > | | > > > 800 ++ | > > > | | > > > 700 ++ ..*...*....*...*. * | > > > 600 *+. .. . | > > > | . .. | > > > 500 ++ . | > > > | * | > > > 400 > > > ++-------------------------------------------------------------------+ > > > > > > > > > iostat.sda.w_await > > > > > > 1200 ++-------------------------------------------------------------------+ > > > | | > > > 1100 O+ O O O O O O O O O O O O O O O > > > 1000 ++ | > > > | | > > > 900 ++ | > > > | | > > > 800 ++ | > > > | | > > > 700 ++ ..*...*....*...*. * | > > > 600 *+. .. . | > > > | . .. | > > > 500 ++ . | > > > | * | > > > 400 > > > ++-------------------------------------------------------------------+ > > > > > > > > > [*] 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, > > > Fengguang