--Jauder
On Thu, 25 Apr 1996, J.J. Burgess wrote:
> Below is a quick crude program to measure swap performance,
> if the swap size is set to 8meg, the program takes about 2 seconds to
> complete, if it is set to 12meg it takes over 2 minutes (with only about
> 2 seconds of user+system) this is with 16meg. of ram. Surely we can
> better this factor of 60 slowdown?
> (This is on a 16meg machine, 1.3.95)
>
> - obviously something isn't doing any large predictive read-ahead swap in
> or page outs!.
>
>
> /* swaptest.c 0.1 (C) Jon Burgess 92jjb@eng.cam.ac.uk
> *
> * Tests the performance of the linux swap system
> * Generates a block of random data, reading it a number of times.
> */
> /* TODO: options
> * <size> memory size to try
> * -time do automatic timing
> * -random random pattern of data access
> * -dirty dirty data on every access
> * -cycles number of iterations of read to try.
> */
>
>
> #include <stdio.h>
> #include <sys/types.h>
> #include <sys/stat.h>
> #include <fcntl.h>
>
> #define MEM_MB 12
> #define MEM_KB (MEM_MB*1024)
> #define MEM_SIZE (MEM_KB*1024)
> #define ITERATIONS 2
>
> main(int argc, char *argv[])
> {
> char *buf;
> int i,ptr;
> char junk;
> FILE *file;
>
> buf = (char *) malloc(MEM_SIZE);
>
> file = fopen ("/dev/urandom","r");
> if (!file) {
> printf("Error reading /dev/random\n");
> exit(1);
> }
>
> /* Get a 1kB of random data */
> if (1 != fread( buf, 1024, 1, file)) {
> printf("Error reading random data\n");
> exit(1);
> }
> for (i=1; i<MEM_KB;i++)
> memcpy(buf+(i*1024),buf,1024);
>
> for(i=0;i<ITERATIONS;i++) {
> printf("Running loop %d\n",i);
> for(ptr=0; ptr<MEM_SIZE; ptr+=4096){
> junk = *(ptr+buf);
>
> /* Do we dirty each page? */
> #if 0
> *(ptr+buf) = 0;
> #endif
> }
> }
> exit(0);
> }
>
>
>
> Thanks.
>
> .. . . . . . . . . . . . . . ..
> :: : : Jon Burgess 01223-461907 : : ::
> :: : jjb1003@cam.ac.uk : : ::
> :: : : : : : : : : : : : : : ::
>
>
.sig under construction