Dear squidder,
I'm currently having performance problem with squid on my Linux 2.0.37p4
(with filedescriptor.patch.linux.v8.04) system. The hardware is PII-400,
512MB Ram, 7x4GB UWscsi (6 for cache), 2xAHA2940UW SCSI Host Adapter,
ASUS-P2B-LS motherboard. It is a dedicate squid server running
squid-2.0.PATCH2.
I recently upgrade this server from Linux 2.0.36pre5 to 2.0.37pre4 and
notice that after running for a while (~1hr) Linux start swapping too
much, see below:
[root@proxy1 kernel]# free
total used free shared buffers cached
Mem: 515808 508124 7684 26340 73364 95324
-/+ buffers/cache: 339436 176372
Swap: 257000 238112 18888
and from cachemgr.cgi:
Memory usage for squid via mallinfo():
Total space in arena: 294640 KB
Ordinary blocks: 175476 KB 1355927 blks
Small blocks: 0 KB 0 blks
Holding blocks: 4564 KB 6 blks
Free Small blocks: 0 KB
Free Ordinary blocks: 119164 KB
Total in use: 180040 KB 61%
Total free: 119164 KB 40%
Memory accounted for:
Total accounted: 145384 KB
If I'm not wrong,
Mem used = squid + cached + buffers + shared + other processes
= 294640 + 95324 + 73364 + 26340 + ~20000
= ~509668 (~mem used)
I've no idea why Linux swap that much. Do you have any suggestion? I think
before linux start swapping it should try to reduce cache or buffer size
to some number (is it configurable?). I never see this behavior with
2.0.36p5 so I think MM policy is change between these versions.
Your suggestion is greatly appreciate.
Regards,
--
aet
