On Wed, 12 May 2010, Andy Dills wrote: > amavisd-new-2.6.4_5,1 > > BTW, as a shot in the dark I built the 8.0-STABLE kernel with "options > DEADLKRES" but that didn't help. I'm now in the process of building the > kernel without SMP support to see how that behaves.
It was the same behavior without SMP, so I went back to the SMP for the obvious performance boost. I even eventually had problems when max_servers was set to 10, so I lowered it to 8 and am letting it run in production to see how things pan out. I'm wondering if my experience is in fact typical of the more current software revisions on fast hardware, and that I might have made a mistake in judging performance based on number of max_servers attainable...sort of a comparing apples and oranges situation. Because I'm able to process a message in under 2 seconds compared to around 5 on the old software/hardware, I may actually be able to handle significantly more messages per hour with many fewer amavisd processes. Regardless, I'm really curious as to why amavisd would behave like this once it hits a certain threshold...it's not like it ever backs off, once it decides it's going to spiral out of control it never recovers until a "reload". It's definitely a concern that I'd like to get ironed out, but I'm feeling better about rolling out this build to the new cluster in the short term. Andy --- Andy Dills Xecunet, Inc. www.xecu.net 301-682-9972 --- ------------------------------------------------------------------------------ _______________________________________________ AMaViS-user mailing list AMaViS-user@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/amavis-user AMaViS-FAQ:http://www.amavis.org/amavis-faq.php3 AMaViS-HowTos:http://www.amavis.org/howto/