I have a strange situation, where I wonder if disabling /swap will help, or solve.
My squid server runs as a http accelerator, it is a HP/Compaq DL360 with
2 Gb of memory, 4 squid process running as it has 2 hyperthread cpus.
It serves ~ 150 req/s (each process), its Linux RH9 kernel 2.4.
After some time running/serving , memory usage looks like this :


            total       used       free     shared    buffers     cached
Mem:       2069488    2059504       9984          0      72164    1466184
-/+ buffers/cache:     521156    1548332
Swap:      2097112     119128    1977984

The thing is that swap usage keeps growing until the machine consumes all the 2G
swap partition. It took weeks to fill up the swap area, but eventually it will be 100%.
I set squid cache_mem to 32 mb because i guessed that this 1.5 cached memory
could only be file cache, as this machine does nothing more than running squid.


14781 nobody    18   0  105M 100M  3180 NYI S    27.1  4.9  1802m squid
22228 nobody    12   0  101M 100M  2192 NYI S     3.8  4.9  1333m squid
22358 nobody    16   0 69148  66M  8248 NYI S    19.4  3.3 399:07 squid
22374 nobody    13   0 69488  66M  3344 NYI S     7.7  3.2 414:42 squid

So each squid process takes no more than 100mb, and I have some redir perl scripts
running, but only takes like 50 mb more. Afterall , this is very odd .
I'm looking forward for any sugestions .


Thanks in advance.
Regards
Bruno Marcondes

Reply via email to