On Thu, 4 Nov 2010 10:09:19 -0400
Wendy J Bossons <wboss...@mit.edu> wrote:
> What is the overhead of opening and closing socket connections to
> clamd?
Depends on system load, if all your CPUs are busy it might take until
the next time slice till clamd gets a chance to respond.
>
> I am trying to get some metric on that, but they are not making sense
> to me.
I'd expect UNIX sockets to be faster than TCP.
Try measuring 'time clamdscan --version'. That connect to clamd,
queries version, and shows reply.
>
> What should be the amount of time in milliseconds that this takes,
I would be worried if it takes a milisecond.
This is on a clamav under heavy load:
connect time:0.052 ms avg, 0.278 ms max
reply time:0.018 ms avg, 0.056 ms max
connect time:0.153 ms avg, 10.064 ms max
reply time:0.019 ms avg, 0.068 ms max
#!/usr/bin/perl
use warnings;
use strict;
use IO::Socket::UNIX;
use Time::HiRes qw(gettimeofday tv_interval);
my ($conn_max, $conn_sum, $reply_max, $reply_sum);
my $n = 100;
my $socket = shift;
$conn_max = 0, $conn_sum = 0, $reply_max = 0, $reply_sum = 0;
for (my $i=0;$i<$n;$i++) {
my $t0 = [gettimeofday];
my $s = IO::Socket::UNIX->new($socket) || die "can't connect to
$socket"; my $t1 = [gettimeofday];
$s->print("PING");
$s->getline;
my $t2 = [gettimeofday];
$s->close;
my $conntime = tv_interval($t0, $t1);
$conn_sum += $conntime;
if ($conntime > $conn_max) {
$conn_max = $conntime;
}
my $replytime = tv_interval($t1, $t2);
$reply_sum += $replytime;
if ($replytime > $reply_max) {
$reply_max = $replytime;
}
}
$reply_sum /= $n;
$conn_sum /= $n;
$conn_sum *= 1000;
$reply_sum *= 1000;
$conn_max *= 1000;
$reply_max *= 1000;
printf "connect time:%.3f ms avg, %.3f ms max\n", $conn_sum, $conn_max;
printf "reply time:%.3f ms avg, %.3f ms max\n", $reply_sum, $reply_max;
Best regards,
--Edwin
_______________________________________________
Help us build a comprehensive ClamAV guide: visit http://wiki.clamav.net
http://www.clamav.net/support/ml
