On Sun, 2008-09-14 at 21:51 +0200, Mike Galbraith wrote:
Turns off all netfilter options except tables, etc.
Since 2.6.22.19-cfs-v24.1 and 2.6.23.17-cfs-v24.1 schedulers are
identical, and these are essentially identical with 2.6.24.7, what I
read from numbers below is that cfs in 2.6.23 was somewhat less than
wonderful for either netperf or tbench, Something happened somewhere
other than the scheduler at 23->24 which cost us some performance, and
another something happened at 26->27. I'll likely go looking again..
and likely regret it again ;-)
Math ain't free is part of it, though apparently not much. For me,
tbench regression 22->27 is ~10%, and netperf regression is ~16%.
Data:
2.6.22.19
Throughput 1250.73 MB/sec 4 procs 1.00
16384 87380 1 1 60.01 111272.55 1.00
16384 87380 1 1 60.00 104689.58
16384 87380 1 1 60.00 110733.05
16384 87380 1 1 60.00 110748.88
2.6.22.19-cfs-v24.1
Throughput 1204.14 MB/sec 4 procs .962
16384 87380 1 1 60.01 101799.85 .929
16384 87380 1 1 60.01 101659.41
16384 87380 1 1 60.01 101628.78
16384 87380 1 1 60.01 101700.53
wakeup granularity = 0 (make scheduler as preempt happy as 2.6.22 is)
Throughput 1213.21 MB/sec 4 procs .970
16384 87380 1 1 60.01 108569.27 .992
16384 87380 1 1 60.01 108541.04
16384 87380 1 1 60.00 108579.63
16384 87380 1 1 60.01 108519.09
2.6.23.17
Throughput 1192.49 MB/sec 4 procs .953
16384 87380 1 1 60.00 91124.67 .866
16384 87380 1 1 60.00 93124.38
16384 87380 1 1 60.01 92249.69
16384 87380 1 1 60.01 91103.12
wakeup granularity = 0
Throughput 1200.46 MB/sec 4 procs .959
16384 87380 1 1 60.01 95987.66 .866
16384 87380 1 1 60.01 92819.98
16384 87380 1 1 60.01 95454.00
16384 87380 1 1 60.01 94834.84
2.6.23.17-cfs-v24.1
Throughput 1242.47 MB/sec 4 procs .993
16384 87380 1 1 60.00 101728.34 .931
16384 87380 1 1 60.00 101930.23
16384 87380 1 1 60.00 101803.15
16384 87380 1 1 60.00 101908.29
wakeup granularity = 0
Throughput 1238.68 MB/sec 4 procs .990
16384 87380 1 1 60.01 105871.52 .969
16384 87380 1 1 60.01 105813.11
16384 87380 1 1 60.01 106106.31
16384 87380 1 1 60.01 106310.20
2.6.24.7
Throughput 1202.49 MB/sec 4 procs .961
16384 87380 1 1 60.00 94643.23 .868
16384 87380 1 1 60.00 94754.37
16384 87380 1 1 60.00 94909.77
16384 87380 1 1 60.00 95457.41
wakeup granularity = 0
Throughput 1204 MB/sec 4 procs .962
16384 87380 1 1 60.00 99599.27 .910
16384 87380 1 1 60.00 99439.95
16384 87380 1 1 60.00 99556.38
16384 87380 1 1 60.00 99500.45
2.6.25.17
Throughput 1220.47 MB/sec 4 procs .975
16384 87380 1 1 60.00 94641.06 .867
16384 87380 1 1 60.00 94864.87
16384 87380 1 1 60.01 95033.81
16384 87380 1 1 60.00 94863.49
wakeup granularity = 0
Throughput 1223.16 MB/sec 4 procs .977
16384 87380 1 1 60.00 101768.95 .930
16384 87380 1 1 60.00 101888.46
16384 87380 1 1 60.01 101608.21
16384 87380 1 1 60.01 101833.05
2.6.26.5
Throughput 1182.24 MB/sec 4 procs .945
16384 87380 1 1 60.00 93814.75 .854
16384 87380 1 1 60.00 94173.41
16384 87380 1 1 60.00 92925.24
16384 87380 1 1 60.00 93002.51
wakeup granularity = 0
Throughput 1183.47 MB/sec 4 procs .945
16384 87380 1 1 60.00 100837.12 .922
16384 87380 1 1 60.00 101230.12
16384 87380 1 1 60.00 100868.45
16384 87380 1 1 60.00 100491.41
2.6.27
Throughput 1088.17 MB/sec 4 procs .870
16384 87380 1 1 60.00 84225.59 .766
16384 87380 1 1 60.00 83362.65
16384 87380 1 1 60.00 84060.73
16384 87380 1 1 60.00 83462.72
wakeup granularity = 0
Throughput 1116.22 MB/sec 4 procs .892
16384 87380 1 1 60.00 92502.44 .841
16384 87380 1 1 60.01 92213.72
16384 87380 1 1 60.00 91445.86
16384 87380 1 1 60.00 91832.84
revert sched weight/asym changes, gran = 0
Throughput 1149.16 MB/sec 4 proc .918
16384 87380 1 1 60.00 94824.92 .868
16384 87380 1 1 60.01 94579.45
16384 87380 1 1 60.01 95284.94
16384 87380 1 1 60.01 95228.22
Weight/asym changes cost ~3%. Mysql+oltp agrees. Preempt happy loads
lose a bit, preempt haters gain a bit. Performance shift.
--
