I am wondering if having cache_dir drives on a RAID controller that has Read/Write cache turned on might cause problems? I'm fairly sure that Squid manages the latency, etc of its cache_dir drives. The drives that my Squids use are all on RAID controllers as single volumes. However I recently found that if I enable Read/Write cache on the cache_dir drives that load on the processor goes off the scale. Could it be that Squid gets such a quick response from the drive that it thinks the drive is super fast and thus slams it, causing it to run out of Read/Write cache and then gets overloaded as the requests backlog? I'm hoping that this would explain a number of my Squid servers that can be very unstable. The cache I'm talking about is in the order of only about 128MB worth, and from 3-4 cache_dir drives.

Peter Smith



Reply via email to