Sunday July 19, 2009

After the surprise interest in the maximum number of processes on a system it seems rude not to try and see how many threads I can squeeze into a single process while I have access to a system where physical memory will not be the limiting factor. The expectation is that this will closely match the number of processes as each thread will have an LWP in the kernel which will in turn consume the segkp.

A slight modification to the forker program:

: exdev.eu FSS 62 $; cat thr_creater.c
#include <stdio.h>
#include <stdlib.h>
#include <sys/types.h>
#include <unistd.h>
#include <thread.h>

int
main(int argc, char **argv)
{
        pid_t pid;
        int count=0;
        while(count < (argc != 2 ? 100 : atoi(argv[1])) &&
            (pid = thr_create(NULL, 0, (void * (*)(void *))pause,
            NULL, THR_DETACHED, NULL)) != -1) {
                if (pid == 0 ) {
                        /* Success, ) */
                        if (count % 1000 == 0)
                                printf("%d\n", count);
                        count++;
                }
        }
        if (pid < 0)
                perror("fork");
        printf("%d\n", count);
        pause();
}

and this time it has to be built as a 64 bit program:

# make "CFLAGS=-m64 -mt" thr_creater
#

Here is how it went:

$; ./thr_creater 1000000        
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
10000
11000
12000
13000
14000
15000
.....
782000
783000
784000

Here things have stopped and for some bizarre reason attaching a debugger to see what is going on does not seem to be a good idea. I had prstat running in another window and it reported:

  2336 cg13442  7158M 7157M cpu73    0    0   1:42:59 1.6% thr_creater/784970

Which is just a few more threads than I got processes (784956) when running in multi user. However at this point the system is pretty much a warm brick as if I exit any process thr_creater hoovers up the process so I can create no more. Fortunately I had realized this would happen and had some sleep(1) processes running so I could pause the thr_creater and then kill one of the sleeps to allow me to run a command:

$; ps -o pid,vsz,rss,nlwp,comm -p 2336
   PID  VSZ  RSS NLWP COMMAND
  2336 7329704 7329248 784972 ./thr_creater

as you can see it managed to get another two threads created since the prstat exited.