Hi,
I use a simple little CPU-Benchmark based on the perl "Benchmark" module
for rough performance comparison. Now I wanted to extend it to run the
same test in parallel with a given number of threads. After I
got totally unusable results, I stripped down the code for investigation:
use Time::HiRes qw( gettimeofday tv_interval );
use Benchmark qw(:all :hireswallclock);
use threads;
my $num_threads=8;
my $secs=10;
my (@bench, @thr, $total);
my $testcode=sub {
my $sum; for (my $i=0; $i < 1000; $i++) { $sum+=$i } };
my $t0 = [gettimeofday];
for (my $i=1; $i <= $num_threads; $i++) {
$thr[$i]= threads->create(sub { countit($secs, $testcode) });
}
for (my $i=1; $i <= $num_threads; $i++) {
$bench[$i]=$thr[$i]->join();
printf "%02d: iter: %d\n", $i, $bench[$i]->iters;
$total += $bench[$i]->iters;
}
printf "%04.1f secs iter: %d\n",
tv_interval($t0,[gettimeofday]), $total;
For $num_threads=1 it works as expected; the higher the number of
threads, the more unpredictable it behaves. Instead of running the code
for 10 secs, there is a tendency that each thread runs
~($secs/$num_threads) seconds (+ occasional runaways that terminate
after a totally unpredictable time).
At 1st I thought there might be some general problem between timers and
threads. With the following instead of Benchmark::countit, I get the
expected results:
sub countit {
my ($secs,$code)=@_;
my $count=0; my $start=time();
while ((time()-$start) < $secs ) { $count++; $code->() }
return $count;
};
Does anybody happen to have any experience with Benchmark.pm. Is it
generally not usable with multiple threads or am I just doing something
wrong? (Or is it maybe system dependent - I tried it on several machines
running Debian on Linux 2.6.26 - 3.2.0 amd64)
Regards,
Peter
--
To unsubscribe, e-mail: beginners-unsubscr...@perl.org
For additional commands, e-mail: beginners-h...@perl.org
http://learn.perl.org/
