Hi, list! I've been searching the mailing list archives the past few days regarding an issue I have with Squid. I've configured it as an httpd accelerator for one of our high-bandwidth sites and it has worked pretty well these couple of months. One issue I've just noticed is that squid seems to be "capping" the outgoing traffic (direct access is much faster). So far, the slowdown seems to be 2/3 of what the host is capable of pushing out (e.g. throughput of 3Mbps instead of 10Mbps). Some pertinent details: - squid is running on a 2x2.8GHz Xeon host, 64-bit RH Linux, and 8GB of RAM. - no delay pools - cache_mem of 32M - one cache_dir of 8GB on SATA drive (ext3, noatime), using aufs (diskd is slightly slower) - max cacheable object size of 100M - cache stores many tiny GIFs, and a couple of large objects (video, audio, etc) The following messages in the archive have been helpful, but none have worked for me: - http://marc.theaimsgroup.com/?l=squid-users&m=113771024329403&w=2 - http://www.squid-cache.org/mail-archive/squid-users/200601/0353.html - http://marc.theaimsgroup.com/?l=squid-users&m=105061905404456&w=2 - http://marc.theaimsgroup.com/?l=squid-users&m=105062051805875&w=2 Some strange behavior I see (on a loaded proxy): 1. When I request the object from proxy and the object is cached, throughput is only at about 3Mbps (local LAN) 2. When I request the object from the proxy with pragma-no-cache, the throughput is about 4-6Mbps 3. When I request the object from the web server hosting the file directly, the throughput is about 10Mbps Another interesting tidbit: the source files are served over NFS, whereas the cache is on local disk, so this is confusing to me. When the proxy is idle, #2 and #3 above are very similar (performing at top speed). There aren't any pagefaults reported by cachemgr, or anything in cache.log that seems to stand out. While performance is more than adequate, I'd like to know what could explain the slowdown/cap. Any help is greatly appreciated. Thanks! - gino ledesma
