Thanks for the clarification. I had noted that Apache, Postgres, etc. could use the other proc, but when I watch top running on the system, I see Zope at the top 95% of the time, the Apache front end putters along at 1-2%, and Postgres blips up to 3% every few minutes. The top process uses a lot more cycles than either of the non-Zope daemons. I don't expect this to change much, which I why I lean towards 1 proc. When I'm looking at systems, does Zope performance correlate well with the pystone benchmark? H ----- Original Message ----- From: "Oliver Bleutgen" <myzope@gmx.net> To: "Howard Hansen" <zope@halfmagic.com> Cc: <zope@zope.org> Sent: Sunday, September 15, 2002 3:15 AM Subject: Re: [Zope] New server recommendations
Howard Hansen wrote:
From recent discussions, it appears that Zope doesn't benefit much from a dual processor system, but it does benefit from higher clock rates and more RAM. So should I invest in a faster single-proc system with a couple of gigs of RAM, or go with a slower dual-proc box?
I wouldn't state it this way. From what I read from various posts, there's a subtle difference to what you said.
- A system which has more daemon processes running than zope (likely) can profit from multiprocessor configs, because the processes will spread across processors.
- There is a problem with mulitproc configs when the zope process wanders between procs (cache coherency and whatnot), so the os having the possibitilty to restrict process to single processors is a big advantage.
cheers, oliver