Hi,
You use about 18 x 15GB, 14.8GB specified in squid.conf per cache_dir. Roughly 260GB of cache_dirs.
According to
http://www.squid-cache.org/Doc/FAQ/FAQ-8.html#ss8.11
you need 2,5GB of RAM for the index only. Didn't you ask questions about squid using all your memory?
Squid is running as a non-root user, there is some reserved space for root only and your hitting that barrier.
Deleting a lot of objects due to changed parameters is a tough job for squid. If you want to keep your disk layout lower the size in your cache_dir setting and disable some of them. Monitor the load and have a look with cachemgr.cgi if squid is actually deleting files - it may take a while before you can notice it with 'du'.
This is what I would do:
- deaktivate all cache_dir which are close to 100%
- deaktivate all cache_dirs on one disk
- repartition it with a single partition & mkreiserfs
- put one cache_dir on it
- repeat with the other disks
- if you are on a big uplink hit ratios shouldnt suffer to much, take your time and let squid refill the cache_dirs.
- be careful with the cache_dir size. with your disks not more than 50% and you don't need to keep more than a week of traffic.
If you don't make any mistakes this can be done with nearly no downtime, couple of restarts and some "stop, squid -z, start" . This can be measured in seconds...
- think about using aufs - I tried it recently - Adam Aube recommended it to me - works fine on a high load squid - but IMHO you should reduce the number of cache_dirs first.
Regards, Hendrik Voigtl�nder
