Marcin, In my case all of the data being sent out was small enough and repetitive enough to be in the Linux filesystem cache. That's where I found the best throughput. I think the typical size of the data items were about 8-30MBytes. It was a regular Linux ext3 filesystem. The machine happens to have been a dual dual-core 64-bit 2Ghz Opteron, although I saw some Intel machines with similar performance per CPU but on those I had only one gigabit network interface and one squid. - Dave On Mon, Oct 06, 2008 at 08:09:17PM +0200, Marcin Mazurek wrote: > Dave Dykstra (dwd@xxxxxxxx) napisa?(a): > > > Meanwhile the '-I' option to squid makes it possible to run multiple > > squids serving the same port on the same machine, so you can make use of > > more CPUs. I've got scripts surrounding squid startups to take > > advantage of that. Let me know if you're interested in having them. > > Currently I run a couple machines using 2 squids each on 2 bonded > > gigabit interfaces in order to get over 200 Mbytes/second throughput. > > > > > What kind of storage do You use for such a IO performance, and what file > system type on it, if that's not a secret:) > > br > > -- > Marcin Mazurek >
