[BENCHMARK] ctxbench - 25% drop in IPC rates

Bill Davidsen (davidsen@tmr.com)
Sun, 29 Dec 2002 20:58:29 -0500 (EST)


In these results, note that the IPC by signal and by SysV message queue
are down by ~25%. Something changed. Everything dropped a little, but
those two really fell off.

As always, "uni" is a single CPU, smp is dual CPU, nosmp is an smp kernel
running on one processor. All smp kernels show more variation between high
and low than uni.

I am considering running with preempt on for these, but I have no historic
data for earlier kernels. Unless it really seems useful, I won't bother
building yet another kernel.

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

Run: 2.5.50nosmp-bl
CPU_MHz 498.049
CPUtype Celeron (Mendocino)
HostName bilbo.tmr.com
KernelName 2.5.50smp
Ncpu 1
Run: 2.5.50smp-bl
CPU_MHz 497.898
CPUtype Celeron (Mendocino)
HostName bilbo.tmr.com
KernelName 2.5.50smp
Ncpu 2
Run: 2.5.50uni-bl
CPU_MHz 497.953
CPUtype Celeron (Mendocino)
HostName bilbo.tmr.com
KernelName 2.5.50
Ncpu 1
Run: 2.5.52nosmp-bl
CPU_MHz 497.929
CPUtype Celeron (Mendocino)
HostName bilbo.tmr.com
KernelName 2.5.52smp
Ncpu 1
Run: 2.5.52smp-bl
CPU_MHz 497.934
CPUtype Celeron (Mendocino)
HostName bilbo.tmr.com
KernelName 2.5.52smp
Ncpu 2
Run: 2.5.52uni-bl
CPU_MHz 497.965
CPUtype Celeron (Mendocino)
HostName bilbo.tmr.com
KernelName 2.5.52
Ncpu 1

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

loops/sec
SIGUSR1 low high average
2.5.50nosmp-bl 51675 56403 54818
2.5.50smp-bl 8841 55534 38632
2.5.50uni-bl 66374 66536 66434
2.5.52nosmp-bl 40230 40322 40264
2.5.52smp-bl 26179 39434 32565
2.5.52uni-bl 49606 50246 49939

loops/sec
message queue low high average
2.5.50nosmp-bl 104965 105261 105105
2.5.50smp-bl 46184 104640 71925
2.5.50uni-bl 121383 122357 121851
2.5.52nosmp-bl 78717 78841 78763
2.5.52smp-bl 47748 63108 53837
2.5.52uni-bl 93991 94909 94352

loops/sec
pipes low high average
2.5.50nosmp-bl 81211 81529 81411
2.5.50smp-bl 42573 70910 56900
2.5.50uni-bl 127910 129669 128603
2.5.52nosmp-bl 76852 77215 77026
2.5.52smp-bl 49571 63707 57279
2.5.52uni-bl 111056 115477 113687

loops/sec
semiphore low high average
2.5.50nosmp-bl 110582 112044 111450
2.5.50smp-bl 54085 78742 63654
2.5.50uni-bl 131425 132274 131794
2.5.52nosmp-bl 102732 103045 102924
2.5.52smp-bl 45872 70347 55434
2.5.52uni-bl 119345 130644 126665

loops/sec
spin+yield low high average
2.5.50nosmp-bl 252055 252529 252361
2.5.50smp-bl 244146 344592 279022
2.5.50uni-bl 322742 325100 323594
2.5.52nosmp-bl 240059 240287 240149
2.5.52smp-bl 359924 435936 389357
2.5.52uni-bl 315797 317316 316650

loops/sec
spinlock low high average
2.5.50nosmp-bl 3 3 3
2.5.50smp-bl 1194785 1194882 1194841
2.5.50uni-bl 3 3 3
2.5.52nosmp-bl 3 3 3
2.5.52smp-bl 1194666 1195555 1195141
2.5.52uni-bl 3 3 3

-- 
bill davidsen, CTO TMR Associates, Inc <davidsen@tmr.com>
  Having the feature freeze for Linux 2.5 on Hallow'een is appropriate,
since using 2.5 kernels includes a lot of things jumping out of dark
corners to scare you.

- 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/