Re: Sarge Performance
On May 22 2005, Lars Roland wrote:
> I have doubled checked these numbers and they are true (perhaps Perl
> 5.8 is much faster than the old version 5.6 on Redhat) - note that
> most people has performance problems with spamassassin because they
> have crappy DNS perfoamnce, I use a dedicated DNS server as a DNS
> cache (djbdns cache) to achive high performance.
I have missed the other e-mails, but if you are benchmarking just one
instance of spamassassin, you may be hitting one "feature" that
spamassassin 3 requires quite a lot more of memory than spamassassin 2 did,
which could mean that, if many instances are running simultaneously, then
you may be hitting swap more requently than with your older setup.
Are you using NFS? If not, then you can try to see if configuring
spamassassin 3 with the option
lock_method flock
gives you any improvement on the situation (see if it is worth using on
your system on the Mail::SpamAssassin::Conf page).
If the problems of swapping are hitting you, then you might get better
performance if you limit the number of concurrent processes.
Also, trying a newer kernel may help with virtual memory considerations
(and tuning the contents of /proc/sys/vm/swappiness may help you here
also).
Hope this helps, Rogério.
P.S.: All this is a shot in the dark, as I got the thread in the middle.
--
Rogério Brito : rbrito@ime.usp.br : http://www.ime.usp.br/~rbrito
Homepage of the algorithms package : http://algorithms.berlios.de
Homepage on freshmeat: http://freshmeat.net/projects/algorithms/
Reply to: