> -----Ursprüngliche Nachricht----- > Von: [EMAIL PROTECTED] > [mailto:[EMAIL PROTECTED] Im Auftrag von Trog > Gesendet: Monday, January 31, 2005 5:22 PM > An: ClamAV users ML > Betreff: RE: [Clamav-users] SESSION BUG - PERFORMANCE? > > > On Mon, 2005-01-31 at 16:06 +0100, Christian Hilgers wrote: > > > > > > > It's a known limitation. > > > > Will this be fixed? > > I expect so, when the protocol is replaced with something else. > > > > > I did some performance testing when using SESSION. > > The first scan is ok but then the performance gets bad. > > > > This is the time (s) for scanning (excluded time for > transfer to socket) > > > > first file: > > 28K: 0.000151 > > > > second file: > > 28K: 0.039568 > > > > > > This is a test with 0 Byte Files > > > > 0.000142 (first) > > 0.032147 (second) > > 0.032142 (third) > > > > Ist this also a known problem? > > That seems to be a result of kernel scheduling or socket handling. > > I can reproduce it, but I can also cancel it out by adding a sleep(1) > before closing the stream socket, i.e. by forcing a schedule > before the > scanning starts.
I guess it is scheduling problem too, but I think it's up to clamd. Maybe it is caused by pthread? I did some further testing: You can put the sleep wherever you want to fix it. A for loop can also cancel it out. A sched_yield() only works with a high CPU load on the system! I tried it with two computers and a "real" socket connection. I got the same results. The second run is slower. Christian _______________________________________________ http://lists.clamav.net/cgi-bin/mailman/listinfo/clamav-users
