On Thu, Apr 24, 2008, Alex Rousskov wrote: > My wording may have been poor. I assume we always use a quad box. Add > SMP support to Squid. Test Squid3 on that quad box with SMP disabled > (that is what I called single-core performance on a quad box). Test with > SMP enabled and configure Squid to use two (out of four) cores. Test > with SMP enabled and configure Squid to use three (out of four) cores. > This will give the 1-2 and 1-3 comparisons I was asking about. > > What I would like to do now is to estimate the difference between those > future results. In other words, what kind of scale should we expect > after 5-10 months worth of work, very roughly?
Uhm, i'd start by benchmarking where your current bottlenecks are. If your bottlenecks are memory-bus related, threading Squid won't help (much). If your bottlenecks are CPU related then theading Squid will help a bit. > > * Ok. Look at the performance of varnish, apache2-thread and memcached. > > memcached is a good one. They get reasonably linear performance up to quad > > core iirc where things like memory transaction rates impose limitations on > > performance. > > Do you have any pointers to varnish, apache2-thread, or memcached > performance related to mutli-core scale? Trying to find some. But people run memcache + varnish on systems and seem to get linear-scalability. Of course, quad-core and quad-CPU are different beasts and I bet some of these large installs have >1 socket.. Adrian -- - Xenion - http://www.xenion.com.au/ - VPS Hosting - Commercial Squid Support - - $25/pm entry-level VPSes w/ capped bandwidth charges available in WA -
