Performance numbers with prefer numa hotness over cache hotness

[Srikar Dronamraju]

Ran 5 runs of autonuma-benchmark
(https://github.com/pholasek/autonuma-benchmark)

KernelVersion: 4.1.0-rc6
	Testcase:         Min         Max         Avg      StdDev
  elapsed_numa01:      533.73      789.12      696.77       86.86
  elapsed_numa02:       22.59       28.15       25.87        1.99
	Testcase:         Min         Max         Avg      StdDev
   system_numa01:      342.78     1655.01     1128.26      479.49
   system_numa02:       36.12       99.35       62.93       20.95
	Testcase:         Min         Max         Avg      StdDev
     user_numa01:    43724.45    62948.12    55832.97     6486.37
     user_numa02:     1693.81     1945.30     1817.80      101.85
	Testcase:         Min         Max         Avg      StdDev
    total_numa01:    44067.20    64603.10    56961.22     6941.77
    total_numa02:     1729.93     2044.65     1880.73      118.16

KernelVersion: 4.1.0-rc6+patch
	Testcase:         Min         Max         Avg      StdDev     %Change
  elapsed_numa01:      484.08      726.45      648.46       85.14       6.65%
  elapsed_numa02:       20.38       29.56       24.99        2.93       2.98%
	Testcase:         Min         Max         Avg      StdDev     %Change
   system_numa01:      325.60     1111.46      837.76      290.98      26.14%
   system_numa02:       47.65       73.59       60.59       10.97       3.18%
	Testcase:         Min         Max         Avg      StdDev     %Change
     user_numa01:    39256.04    53387.31    48009.88     4717.79      14.65%
     user_numa02:     1498.92     2089.51     1762.66      193.30       2.64%
	Testcase:         Min         Max         Avg      StdDev     %Change
    total_numa01:    39581.60    54483.60    48847.66     4987.34      14.89%
    total_numa02:     1546.57     2161.73     1823.26      202.28       2.66%

Performance counter stats for 'system wide': (for 1st run).
numa01
(Before patch)
          5,81,911      cs                                         [100.00%]
          1,07,380      migrations                                 [100.00%]
          5,15,465      faults
    2,99,19,83,196      cache-misses

     724.330728328 seconds time elapsed

(After patch)
          3,04,022      cs                                         [100.00%]
            47,539      migrations                                 [100.00%]
          3,14,508      faults
    1,91,03,93,197      cache-misses

     484.088557526 seconds time elapsed

numa02
(Before patch)
            26,078      cs                                         [100.00%]
             4,932      migrations                                 [100.00%]
            28,025      faults
      11,75,46,393      cache-misses

      28.156065587 seconds time elapsed

(After patch)
            18,972      cs                                         [100.00%]
             3,650      migrations                                 [100.00%]
            24,620      fault
       9,19,64,503      cache-misses

      20.382626292 seconds time elapsed

# numactl -H
available: 4 nodes (0-3)
node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
node 0 size: 32425 MB
node 0 free: 25493 MB
node 1 cpus: 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47
node 1 size: 31711 MB
node 1 free: 30933 MB
node 2 cpus: 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71
node 2 size: 30431 MB
node 2 free: 29577 MB
node 3 cpus: 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95
node 3 size: 32219 MB
node 3 free: 31455 MB
node distances:
node   0   1   2   3
  0:  10  20  40  40
  1:  20  10  40  40
  2:  40  40  10  20
  3:  40  40  20  10

Srikar Dronamraju (1):
  sched:Prefer numa hotness over cache hotness

 kernel/sched/fair.c     | 96 ++++++++++++++++++-------------------------------
 kernel/sched/features.h | 18 +++-------
 2 files changed, 40 insertions(+), 74 deletions(-)

--
1.8.3.1

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/