On Mon, Jan 28, 2008, Chris Woodfield wrote: > This does bring an interesting question - is it possible to give squid > *too much* memory? > > My theoretical setup would be an uber-box (32GB RAM, multi-TB of disk) > running 64-bit squid and with mem_cache set to something in the > 25-30GB range (as high as we can without swap risk), with a > maximum_object_size_in_memory in the multiple MB; we want to > effectively cache as much as possible in memory as opposed to disk. > Squid and associated utilities will be the only thing running on the > box. > > Does this make sense, or is a more balanced approach re: squid > cache_mem vs. kernel page cache allocation going to provide better > performance? Squid's great at keeping small objects in memory, but not large objects. Its been a known problem for a while. I've got a dotpoint to fix it in Squid-2.X; 3.x has a different memory object cache organisation which will need to be revisited once I've got some more experience with memory caching in 2.x. Adrian
