Scaling Zope Presentation
"Scaling Zope" presentation online.
Scaling Zope is a presentation given at the 2003 Plone conference.
multiple squid tiers
I finally watched the video today
(see http://plone.org/events/conferences/1/video/index_html#scaling )
and there's one thing I'd always wondered about that is answered in a q&a somewhere in the
middle of the talk: Why two tiers of Squid caches?
The answer is that the first (outer) tier handles all incoming requests from the outward-facing load balancer... no surprise there. This tier in turn makes requests to the second (inner) squid tier which is smaller (fewer machines). The small inner tier has the effect of reducing the number of systems that are making requests to the second load balancer that sits between the squid cluster and the zope/zeo cluster. The end result is reduced load on the zope/zeo cluster.
(see http://plone.org/events/conferences/1/video/index_html#scaling )
and there's one thing I'd always wondered about that is answered in a q&a somewhere in the
middle of the talk: Why two tiers of Squid caches?
The answer is that the first (outer) tier handles all incoming requests from the outward-facing load balancer... no surprise there. This tier in turn makes requests to the second (inner) squid tier which is smaller (fewer machines). The small inner tier has the effect of reducing the number of systems that are making requests to the second load balancer that sits between the squid cluster and the zope/zeo cluster. The end result is reduced load on the zope/zeo cluster.
Great presentation....
I have played around with a ton of configurations looking for the best mix of interval checks, threads and cache.
I generally run about 10 threads, with a nice of -10, 5000 objects per cache and an interval I think at 2000. I also RAM cache some bigger dtml methods that return the same thing over and over.
So far this has worked pretty darn well.
Anyway, great presentation.
BZ