Hi > For now, the problems are: a) they do not perform very well in peak times. > Because they needed much too much RAM, I limited the cache_dir to only a few > 100 MB, even if we have over 20 Gigs in each. Still, in the evening they get > quite slow. Well, they ARE quite loaded for sure... Did you compile Squid with './configure --enable-async-io'? You should, if your caches are loaded. Check the FAQ for more info. > The other question is a little linux 2.2 specific: It's stated that writing > a higher value in file-max in /proc/sys/kernel or something gives you more > filehandles. But I can do what I want, I always have 1024. That's enough for > the moment, but I get I'll get short on them sometimes again... With the > common patches for 2.0.x I could easily get 4096 file descriptors, if I > wanted to. With Alan Cox's patches, 2.2 can have 1024*1024 fds per process Alan's patches are available from: ftp://ftp.*.kernel.org/pub/linux/kernel/alan/ Substitute your country-code for the '*' (or replace it with 'us' if you don't have a local kernel mirror.) Oskar
