Hi folks, we're using Squid in an accelerator setup for some time now and it ran very smoothly. But now we're getting into some performance trouble. But before buying new iron I want to try to squeeze some more performance out of our current setup. So if you have any hints on doing this please share them. Here the facts: - Version: 2.7.STABLE7-2 - triple sibling peers with two parents - using acl urlpath_regex since data are partially different on parents - max Traffic: 30 MBit/s Outgoing - 8 MBit/s ingoing - max 800 req/s - 2x Xeon Dualcore 2 GHz - 4x 146 GB disks - max CPU Usage: Total: 24% Wait: 16% User: 6% System: 2% Configuration: I post only the important parts(hopefully...) cache_peer server1 parent 80 0 no-query no-digest originserver round-robin name=s1 cache_peer server2 parent 80 0 no-query no-digest originserver round-robin name=s2 # Siblings exchange cache digest cache_peer image10 sibling 80 0 proxy-only name=img10 cache_peer image45 sibling 80 0 proxy-only name=img45 # we have 16GB and use 8 as cache cache_mem 8192 MB # Mean Object Size is ca. 28 KB so we use 50 KB maximum_object_size_in_memory 51200 bytes memory_replacement_policy lru cache_replacement_policy lru store_dir_select_algorithm least-load collapsed_forwarding on # since our content changes very fast we rebuild the digest every ten minutes digest_rebuild_period 10 minutes digest_rewrite_period 10 minutes # ca. 80 GB per Disk on Reiser-FS cache_dir aufs /web/cache/1 81920 290 256 cache_dir aufs /web/cache/2 81920 290 256 cache_dir aufs /web/cache/3 81920 290 256 cache_dir aufs /web/cache/4 81920 290 256 Since we cache only tons of small pictures I was thinking about implementing COSS. But it's hard to find useful information about it. So if this is a real alternative here, can you please point me to some documents which help with implementing this. I know about [1] but I don't know if this is still valid for 2.7 and it's not easy to wrap my head around it. [1] http://wiki.squid-cache.org/Features/CyclicObjectStorageSystem Cheers, Markus
