Hi there,
I've been looking around for performance issues and possible enhancement of generic squid configuration. We (in our company) are running a linux box with Squid (2.7Stable1, I try to stay up-to-date). The box is a PIII-1GHz with 512Mb of RAM and 3x18Gb hard-drives. Web access is provided by a 10Mb dsl line (download speed. Upload speed is 768Kb) The box handles about 250 clients. Monitoring shows around 1200/1600 HTTP requests (5min average) and 400 HTTP hits (25% hits) DSL line is used between 1.5 and 2 Mb/s. Beside the fact that the hit rate is low, response time are way too long for users (cache-misses median service times are around 200ms and cache-hits are around 3ms) I made changes to the cache_dir : - lowered it from 10240Mb to 1024; - moved from aufs to diskd - moved from diskd to coss - added a second cache disk : -- run both with aufs -- run both with diskd -- run one with diskd and the second with coss I tried the null storage too, in order to check if disk I/O is involved in this situation. Then I tried to move squid to a "newer" box with two 2.4GHz CPUs, 2Gb of RAM and 5 300Gb SCSI disks. The configuration run with 1 system disk and 4 cache disks with diskd. (1024Mb of cache) Tried the null storage again ... Responses times are not better, in any case. Well, in fact, I got a *tiny* improvement when I installed CACHEBOY 1.0 (sorry Henrik, but I had to give it a try ...) As it does not seem to be a squid issue, I thought of modifying the kernel, particulary the NR_OPEN and NR_FILE, updating these values from the original 1024 to 8192. Are these the right values to modify ? Should I dig somewhere else in the system ? Many thanks for your help, Ionel