Performance test of kernel (TICKS test)

zzaappp · 11-10-2008, 03:48 PM

I have a performance test I've been running against a collection of kernels in order to find the fastest commercially provided distro out there. So far the fastest I've found has been the Gentoo distro, but that one can take a lot of effort to set up on new hardware.

I write a TICKS test to examine how fast and accurate the kernel is at waking up processes from a sleeping state.

Here is the code to ticks.c:

Code:

#include <stdlib.h>
#include <stdio.h>
#include <string.h>

#include <sys/time.h>
#include <sys/timeb.h>
#include <unistd.h>
#include <sys/types.h>
#include <sys/times.h>
#include <signal.h>

#include <errno.h>

unsigned long milsec(void)
{
  struct timeval  tv;
  struct timezone tz;

  if( gettimeofday( &tv,&tz ) == -1 )
    return( 0 );
  return( tv.tv_usec + (tv.tv_sec*1000*1000));
}

const char* uname(void)
{
  static char  buff[1024];
  FILE*        fp=fopen("/proc/version","r");

  if( fp==NULL )
    return( "(OS unknown)" );
  fgets(buff,sizeof(buff),fp);
  return( buff );
}

int main(int argc,char *argv[])
{
  volatile char           B1[102400];
  volatile char           B2[102400];
  volatile unsigned long  Start,Stop;
  volatile int            count,max=1000,index;
  struct timeval          tv;
  struct timespec         tvp;

  if( argc >=2 ) 
    max = atoi(argv[1]);

  printf("OS:  %s\n",uname());
  Start = milsec();
  for(count=0; count<max; ++count)
  {
    tv.tv_sec  = 0;
    tv.tv_usec = 1;
    select(0,NULL,NULL,NULL,&tv);
  }
  Stop  = milsec();
  printf("TICK  :  Iterations[%d] Delta[%10lu] Avg[%7.3f]\n",max,Stop-Start,(double)(Stop-Start)/(double)(max));

  Start = milsec();
  for(count=0; count<max; ++count)
  {
    tvp.tv_sec  = 0;
    tvp.tv_nsec = 1;
    pselect(0,NULL,NULL,NULL,&tvp,NULL);
  }
  Stop  = milsec();
  printf("TICKp :  Iterations[%d] Delta[%10lu] Avg[%7.3f]\n",max,Stop-Start,(double)(Stop-Start)/(double)(max));

  ////////////////////////////////////////
  // BENCH of memcpy/indexing
  ////////////////////////////////////////
  
  Start = milsec();
  for(count=0; count<max; ++count)
  {
    for(index=0; index<sizeof(B1); ++index)
      B2[index] = B1[index];
  }
  Stop  = milsec();
  printf("MEMCPY:  Iterations[%d] Delta[%10lu] Avg[%7.3f]\n",max,Stop-Start,(double)(Stop-Start)/(double)(max));
}

If you save the above code snippet into the file, ticks.c, you can then build and run it with:

Code:

$ gcc ticks.c
$ a.out

I was hoping some REDHAT folks out there could run this and add your results to this thread so I can know how the different permutations of the redhat distro perform. To start, here are my Gentoo stats:

Quote:

Linux foghorn 2.6.20-gentoo-r8 #1 Tue Jul 17 10:48:20 EDT 2007 x86_64 AMD Athlon(tm) 64 FX-55 Processor AuthenticAMD GNU/Linux
TICKS : Iterations[1000] Delta[ 255] Avg[ 0.255]

I really appreciate any assistance with this!

-z

jailbait · 11-10-2008, 03:57 PM

How do you factor out differences in CPU speed from one computer to another?

-----------------
Steve Stites

zzaappp · 11-10-2008, 04:00 PM

Just run the test and post results. The kernel will not be under a load: This is simply a knee-jerk response to tell me something about the patches that have been applied to the kernel beyond the kernel.org base kernel.

unSpawn · 11-10-2008, 04:08 PM

And please don't crosspost again. Thanks.

T3slider · 11-10-2008, 05:28 PM

Quote:

Originally Posted by zzaappp

Just run the test and post results. The kernel will not be under a load: This is simply a knee-jerk response to tell me something about the patches that have been applied to the kernel beyond the kernel.org base kernel.

Slackware is very simple and transparent. If you want to see what patches were applied, just check the source -- there are SlackBuilds abound for kernel compiling (though they are more complex than your average SlackBuild, and I haven't gone through them). As far as I know, the only significant patch applied is here.

rkrishna · 11-11-2008, 01:33 AM

Code:

bash-3.1$ ./a.out 
OS:  Linux version 2.6.23.9 (root@rkrishna) (gcc version 4.1.2) #2 Wed Dec 12 16:40:16 IST 2007

TICK  :  Iterations[1000] Delta[   4023606] Avg[4023.606]
TICKp :  Iterations[1000] Delta[   3999925] Avg[3999.925]
MEMCPY:  Iterations[1000] Delta[    862911] Avg[862.911]

H_TeXMeX_H · 11-11-2008, 06:19 AM

I'm not sure what this test is supposed to measure, but here's my data on slamd64:

Code:

bash-3.1$ ./ticks
OS:  Linux version 2.6.24.7 (kafox@demonslayer) (gcc version 4.2.3) #1 SMP Sat Sep 20 18:44:11 EEST 2008

TICK  :  Iterations[1000] Delta[    999915] Avg[999.915]
TICKp :  Iterations[1000] Delta[    999974] Avg[999.974]
MEMCPY:  Iterations[1000] Delta[    407857] Avg[407.857]

Doesn't the processor make a difference ? I have a 'Intel(R) Core(TM)2 Quad CPU Q9300 @ 2.50GHz'

fedoralinuxjunkie · 12-09-2008, 02:14 PM

Code:

randy@linux-gx42:~> ./a.out
OS:  Linux version 2.6.25.18-0.2-pae (geeko@buildhost) (gcc version 4.3.1 20080507 (prerelease) [gcc-4_3-branch revision 135036] (SUSE Linux) ) #1 SMP 2008-10-21 16:30:26 +0200

TICK  :  Iterations[1000] Delta[   4009320] Avg[4009.320]
TICKp :  Iterations[1000] Delta[   3999248] Avg[3999.248]
MEMCPY:  Iterations[1000] Delta[   1442285] Avg[1442.285]

Running OpenSuSE 11.0 32-bit, AMD Sempron LE-1300 processor.