Skip to contentSkip to conversation

incorrect time accouting

DISCUSSION Peter Chubb 2 participants 2 messages Sep 3, 2004 1 message not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

#1 lore raw
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

"Paolo" == Paolo Molaro <[email protected]> writes:

Paolo> While benchmarking, a user pointed out that time(1) reported Paolo> incorrect user and system times when running mono. Paolo> A typical example (running on 2.6.8.1 is):

This is because mono is multithreaded. At present, time(1) records only the time of the parent thread. This will change soon (I hope) when the Roland McGrath's getrusage() patches are merged.

--
Dr Peter Chubb http://www.gelato.unsw.edu.au peterc AT gelato.unsw.edu.au The technical we do immediately, the political takes *forever*

#2 lore raw
Quoted 7 lines · On 09/03/04 Peter Chubb wrote:

Paolo> While benchmarking, a user pointed out that time(1) reported Paolo> incorrect user and system times when running mono. Paolo> A typical example (running on 2.6.8.1 is):

This is because mono is multithreaded. At present, time(1) records only the time of the parent thread. This will change soon (I hope) when the Roland McGrath's getrusage() patches are merged.

Thanks for the feedback. This doesn't explain, though, why on 2.4.x and 2.2.x sometimes sensible results are reported and sometimes not and it doesn't explain the results I got with a simple pthread test case on 2.6.8.1. [...doing more testing...]
See the attached test case: the results reported are:

$ gcc test-thread-rusage.c -lpthread
$ time ./a.out join
using a subthread
done

real 0m1.469s
user 0m1.461s
sys 0m0.003s

$ time ./a.out sleep
using a subthread

real 0m1.007s
user 0m0.000s
sys 0m0.001s

So it looks like times for subthreads are accounted for, but only when the thread is joined (or if it is given time to properly exit). I guess Roland's patch takes care of the case when a subthread is destroyed by exiting from main() as well?
A little tracing shows that the intermittent 2.4/2.2 results and the always low timings on 2.6 are probably due to the main program exiting before the subthread cleaned up after itself (the results are intermittent because it's a timing issue: we just wait for the subthread to finish it's work and it's preempted away before it has a chance of calling pthread_exit()).
Thanks.

lupus

Signature · 3 lines
-----------------------------------------------------------------
[email protected]                                     debian/rules
[email protected]                             Monkeys do it better

#include <sys/time.h>
#include <sys/resource.h>
#include <unistd.h>
#include <pthread.h>

void* do_the_work (void *data) {
	int *v = data;
	int i;
	for (i = 0; i < 300000000; ++i) {
		*v = *v + 1;
	}
	printf ("done\n");
	return NULL;
}

int

main (int argc, char *argv[]) {
	int v = 0;
	pthread_t thread;
	if (argc > 1) {
		printf ("using a subthread\n");
		pthread_create (&thread, NULL, do_the_work, &v);
		/* the loop above takes more than 1 second on my box, 
		 * so this exits before the subthread has finished its work.
		 */
		if (strcmp (argv [1], "sleep") == 0)
			sleep (1);
		else if (strcmp (argv [1], "join") == 0)
			pthread_join (thread, NULL);
	} else {
		printf ("doing the work in main()\n");
		do_the_work (&v);
	}
	return 0;
}