In article <5bi10r$un6$1@bilbo.ntc.nokia.com>,
Jari Kokko  <jkokko@cpcemail.ntc.nokia.com> wrote:
> Well, I have the options at 7, and it seems to work for me... Mind
> you, the machine is under little load and memory use most of the time.
> Swap is at 25% or so mostly. I only get hit by the everything-dumps
> bug when I really use swap, like running xv -wait 1 for dozens of big
> images. Usually, some big process, like emacs, when used after that is
> in a non-deterministic state, but nothing else is affected, except
> rarely, when everything starts dumping core.

Setting options to 5 removed the bug on my system completely. I
exercised the kernel swapper by running two 10MB+ xv's and an
allocator program the same time. The system survived just fine.

Thanks for the emailed suggestion. I had tried options at 1 3 and 7
before, but not 5, because if the comment that says that interesting
values are 1, 3, 7.

Jari Kokko

PS. Short allocator program source follows (use e.g.'limit 10m' before
running it!)

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

typedef struct { char megabyte[1024*1024]; } Meg;

main()
{
        Meg *megs[10240];
        static Meg onemeg;
        int i;
        
        fprintf(stderr, "Each `.' means %d bytes.\n", sizeof (onemeg));
        
        fputs("allocating\t", stderr);
        for (i=0 ; ; i++) {
                megs[i]=malloc(sizeof (onemeg)); 
                if (megs[i] == 0) break;
                putc('.', stderr);
        }
        
        fputs("\ncopying\t\t", stderr);
        for (i=0; megs[i]; i++) {
                memcpy(megs[i], &onemeg, sizeof (onemeg)); 
                putc('.', stderr);
        }
        
        fputs("\nfreeing\t\t", stderr);
        for (i=0; megs[i]; i++) {
                free(megs[i]); 
                putc('.', stderr);
        }
        putchar('\n');
}