On Wed, Jul 21, 2010 at 4:57 PM, Marcus Kool <marcus.kool@xxxxxxxxxxxxxxx> wrote: > yes. > 1) the index is in memory and needs 10-20 MB index in memory for each GB on disk I was under the impression (from the oriely squid manual) that recent versions do not use up extra RAM with bigger caches. But maybe I read it wrong? Also, I'm a bit confused as there's only one apparent "memory" option in the squid configuration. Could you explain/point me to a tutorial? > 2) the housekeeping of the index costs more CPU cycles for a larger cache > 3) the housekeeping of the cached objects on disk costs time and grows when the cache is larger. Can be minimised by having cache_swap_low 92 and cache_swap_high 93. > > The system has 2 GB memory, assuming that the system is dedicated for Squid > you need 400 MB for the OS, leaving 1.6 GB for Squid. > A safe value for cache_mem would be 500 MB > > There are many tuning parameters. > The best one is to have more disks. > > Marcus > > Marcello Romani wrote: >> >> Ralf Hildebrandt ha scritto: >>> >>> * Ralf Hildebrandt <Ralf.Hildebrandt@xxxxxxxxxx>: >>> >>>> maximum_object_size 50 MB >>>> cache_dir diskd /squid-cache 45000 16 16 >>>> request_header_max_size 15 KB >>>> request_body_max_size 750 MB >>>> >>>> The machine is 32 bits, MemTotal: 2060960 kB >>> >>> Some stats from before "the purge": >>> >>> 1.4Mio cached objects >>> 42GB Cache size >>> >> >> Could it be that cache_mem + memory required to manage 42GB of cache caused the squid process to be swapped ? >> >