The run queue length (or "load average") is an estimate of processes
that *could* run if there were CPU cycles available. However,
there's a fudge factor because the kernel also figures in processes
that it *expects* to be runnable within the next few seconds
("short-term sleepers"); e.g., processes waiting for I/O or fork()s
to complete. Problems can arise when some processes treat this
number as TRVTH. For example, sendmail refuses connections if
the load average goes above certain levels.
Below is a cute program that I've used to pump up load averages.
Interestingly, FreeBSD is clever enough to avoid figuring such a
"vfork() chain" into its load average computations.
Regards,
Romain Kang Disclaimer: I speak for myself alone,
[EMAIL PROTECTED] except when indicated otherwise.
main(argc, argv)
int argc;
char **argv;
{
int i, nproc;
int rc, wc;
if (argc < 2) {
exit(1);
}
nproc = atoi(argv[1]);
do {
rc = vfork();
if (rc == -1) {
printf("rc = %d, nproc = %d\n", rc, nproc);
pause();
}
else if (rc == 0) {
if (nproc-- > 0)
continue;
}
else {
wait(&wc);
}
} while (nproc > 0);
pause(); /* wait for control-C or whatever */
}
_______________________________________________
Redhat-list mailing list
[EMAIL PROTECTED]
https://listman.redhat.com/mailman/listinfo/redhat-list
