Why this many L1 directories? 96 is a total overkill and may negatively affect performance in garbage collection.
You may actually see increased performance for Squid if you downgrade the boxes to 1 CPU and a non-SMP kernel. Squid cannot effectively use more than one CPU, and running a SMP enabled kernel adds a bit of overhead to many operations, but the difference should not be very big. Alternatively you might consider running two Squid instances on the same server. The areas of tuning you need to look into is primarily * SHM and IPC for diskd. See the Squid FAQ for the Squid requirements. * Memory limits, to make the kernel allow for large processes in the range of 1-1.5GB process size (you do not actually need support for such large processes with this cache configuration, but it is always good to have a little room to grow in). * TCP/IP, primarily to make sure the range of unbound local ports is sufficiently large. Some OS:es defaults to a very small range of unbound ports.. Then you might also want to look into refresh_pattern tuning of Squid to increase the hit ratio. Regards Henrik "Blaser, Shane" wrote: > > Hi All, > > I would like some help to quickly get to the best squid config. Best being > the mix cache performance and box utilization. I have the std squid.conf > with following cache dir setup > > cache_dir diskd /var/squid/0/cache0 10240 96 256 > cache_dir diskd /var/squid/0/cache1 10240 96 256 > cache_dir diskd /var/squid/1/cache0 10240 96 256 > cache_dir diskd /var/squid/1/cache1 10240 96 256 > cache_dir diskd /var/squid/2/cache0 10240 96 256 > cache_dir diskd /var/squid/2/cache1 10240 96 256 > > This is what the disk looks like > > /dev/aacd0s3e 34265868 2916034 28608566 9% /var/squid/0 > /dev/aacd0s3f 34265868 2916032 28608568 9% /var/squid/1 > /dev/aacd0s3g 34266832 2916198 28609288 9% /var/squid/2 > > The boxes run BSD and have 2 processors (2+ gig) with 3 gig of ram > > Any recommendations would be much appreciated .. > > What perf patches are worth trying ??? > > Thanks > > Shane > > .
