seems as if rqtp.tv_nsec = 4 * 1000 * 1000; /* specified in nanoseconds. */ is much better
At 04.11.2003 18:55 +0000, you wrote:
At 31.10.2003 14:34 +0100, you wrote:Hi.
I just put up 1.5b9. This should solve the 'illegal content-length header' problem. Also, starting from this release, we no longer try to accept() any new clients if there is no connection left to assign them to. That means that incoming requests will hang in the kernel, but that is ok I guess. At least that is probably better than accepting and immediately dropping a connection.
Cheers Michiel
sorry, cpu bug (again). the bug of cpu consumption during poll that i corrected in phpstub is in the source code of mathopd again:
to correct it, change function run_children in stub.c to
int run_children(void) { struct pipe_params *p;
struct timespec rqtp; struct timespec rmtp;
rqtp.tv_sec = 0; rqtp.tv_nsec = 4 * 1000 * 1000; /* specified in nanoseconds. */
p = children; while (p) { if (p->cn) { pipe_run(p); nanosleep(&rqtp,&rmtp); } p = p->next; } return 0; }
this little 'slowing down' in fact speeds up the process by reducing the cpu load. although, i didn't have the time to test and tune it completely.
Regards, Thomas
