2009/1/22 Amos Jeffries <squid3@xxxxxxxxxxxxx>: >> How intensive is intensive? At the moment squid is averaging a mere 2.4% >> processor time. > > IIRC older Squid-2 had to step a linked-list the length of the object in 4KB > chunks to perform one of the basic operations (network write I think). Yeah - the memory cache in Squid-2 was really only initially designed as a sort of data pipeline between the server, the store, and the client-side. It sort of "grew" the stuff needed to be a memory cache by virtue, iirc, of wanting to support one incoming stream -> multiple client retrievals without having to always go via the disk store for it. Unless you need the extra boost it gives you in very specific circumstances: * use low cache_mem; but if you notice that you're hitting the disk often; * use a larger cache_mem; but keep maximum_object_size_in_memory down to around 64k Squid-3 "sort of" fixed this. It wasn't ever fully fixed, much like how the problem could be fixed in Squid-2 if someone wanted to do the slight trickery required. Adrian
