[BENCHMARK] IPC performance of recent kernels (ctxbench)

From: Bill Davidsen (davidsen@tmr.com)
Date: Fri Jan 10 2003 - 17:16:16 EST


Sad to say, the IPC rates have been going down since 2.4.18, and I suspect
that will show up in the performance of threaded and cooperating
applications. To some extent the new pthread will help some of this, but
comparing results with a new library vs. old is not useful for measuring
what the kernel is doing.

The column avg/MHz is used for comparing the same system at several CPU
speeds, and getting some feel for P-II, P-II, Athlon and P-IV performance
normalized to clock speed. These test are all on one system at the same
speed, so it's just a number. Likewise the spinlock test is not useful on
uni machines and can be ignored here.

I will have similar results on other systems as I get the chance.

Config descriptions and results follow.

================================================================
    Run information
================================================================

Run: 2.4.18-5.out
  CPU_MHz 348.489
  CPUtype Pentium II (Deschutes)
  HostName oddball.prodigy.com
  KernelName 2.4.18-5
  Ncpu 1
Run: 2.4.20.out
  CPU_MHz 348.492
  CPUtype Pentium II (Deschutes)
  HostName oddball.prodigy.com
  KernelName 2.4.20
  Ncpu 1
Run: 2.5.52.out
  CPU_MHz 348.395
  CPUtype Pentium II (Deschutes)
  HostName oddball.prodigy.com
  KernelName 2.5.52
  Ncpu 1
Run: 2.5.53.out
  CPU_MHz 348.404
  CPUtype Pentium II (Deschutes)
  HostName oddball.prodigy.com
  KernelName 2.5.53
  Ncpu 1
Run: 2.5.54-mm3-1.out
  CPU_MHz 348.404
  CPUtype Pentium II (Deschutes)
  HostName oddball.prodigy.com
  KernelName 2.5.54-mm3
  Ncpu 1
Run: 2.5.54.out
  CPU_MHz 348.444
  CPUtype Pentium II (Deschutes)
  HostName oddball.prodigy.com
  KernelName 2.5.54
  Ncpu 1

================================================================
    Results by IPC type
================================================================

                                   loops/sec
SIGUSR1 low high average avg/MHz
  2.4.18-5.out 46246 47550 46728 134.087
  2.4.20.out 47304 47514 47424 136.083
  2.5.52.out 40961 41699 41290 118.517
  2.5.53.out 40567 41995 41434 118.926
  2.5.54-mm3-1.out 39406 39510 39449 113.228
  2.5.54.out 39573 39699 39647 113.785

                                   loops/sec
message queue low high average avg/MHz
  2.4.18-5.out 89631 90591 90027 258.336
  2.4.20.out 75206 83700 80822 231.920
  2.5.52.out 78427 79163 78675 225.823
  2.5.53.out 75709 75848 75759 217.448
  2.5.54-mm3-1.out 76276 76795 76598 219.854
  2.5.54.out 74131 75390 74969 215.156

                                   loops/sec
pipes low high average avg/MHz
  2.4.18-5.out 86181 95386 91954 263.865
  2.4.20.out 87640 88381 87889 252.198
  2.5.52.out 77020 77586 77365 222.061
  2.5.53.out 67799 68893 68310 196.065
  2.5.54-mm3-1.out 73346 73981 73632 211.343
  2.5.54.out 70445 70572 70489 202.297

                                   loops/sec
semiphore low high average avg/MHz
  2.4.18-5.out 98259 98479 98383 282.315
  2.4.20.out 92052 92569 92307 264.876
  2.5.52.out 84367 84639 84516 242.589
  2.5.53.out 77355 77724 77540 222.560
  2.5.54-mm3-1.out 81461 81732 81552 234.075
  2.5.54.out 77304 77521 77379 222.072

                                   loops/sec
spin+yield low high average avg/MHz
  2.4.18-5.out 200889 203649 201999 579.644
  2.4.20.out 211801 211957 211855 607.921
  2.5.52.out 219279 219658 219470 629.947
  2.5.53.out 180547 180625 180584 518.319
  2.5.54-mm3-1.out 179318 179818 179519 515.261
  2.5.54.out 160157 178830 172598 495.341

                                   loops/sec
spinlock low high average avg/MHz
  2.4.18-5.out 3 3 3 0.010
  2.4.20.out 4 4 4 0.014
  2.5.52.out 3 3 3 0.009
  2.5.53.out 3 3 3 0.009
  2.5.54-mm3-1.out 3 3 3 0.010
  2.5.54.out 3 3 3 0.009

-- 
bill davidsen <davidsen@tmr.com>

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



This archive was generated by hypermail 2b29 : Wed Jan 15 2003 - 22:00:35 EST