On Thu, 12 Feb 2004, Chris Wilcox wrote: > I'm interested in the same question! How did you optimise Squid to get such > an improvement?
things one need to look into * Disk subsystem. Use more than one drive and one of the async cache_dir types. One cache_dir per physical drive, maybe more if using diskd. * If using diskd, remember to carefully read the Squid FAQ on how to configure the OS to support diskd. * Number of filedescriptors. The default of 1024 is not sufficient for high loads. * Total number of sockets, file handles etc allowed to be open in the system, and per-process limits of the same. * SYN backlog size in the OS settings. Especially if you have WAN or Dialup users connecting. * Unbound TCP port range available for outgoing connections. Some systems default with a range of only 4000 ports which quickly run out when approaching 150-200 req/s. * Sufficient amount of memory, and in some cases OS tuning to allow for large processes. Some OS:es also require the swap to be disabled to prevent swapping even if there is sufficient memory. Then monitor the system and tweak things until you see desired results. Regards Henrik
