Henrik Nordstrom wrote:
On ons, 2008-07-16 at 13:20 -0500, Rhino wrote:
Have Squid2.7STABLE1 running on Debian Linux box with 4 278 GB drives
allotted for caching.
System has 32 gig RAM, and serves approx. 10k users.
Seeking input as to the most effective config options on this system in
order to reduce latency
and maximize throughput.
Ample amounts of filedescriptors. Give Squid 32K filedescriptors or so
(--max-fd configure option, or max_filedescriptors config option in
squid.conf 2.7)
maxed number of outgoing TCP ports. (ip_local_port_range sysctl)
persistent connections enabled (default)
aufs cache_dir type, with two cache_dir per drive. L1=128, L2=256.
Make sure to leave some GB of ram for OS + fs cache. Don't be too
agressive about the Squid memory usage.
Do not configure a swap. Instead leave sufficient margin in the memory
usage..
Should also give some advice on which filesystem to use, but ext3 is
probably fine. reiserfs is also a good candidate but may need the notail
mount option.
noatime mount option on the cache drives.
Regards
Henrik
THanks much for the quick response, Henrik.
Filesystem for cache disks currently configured for reiserfs with
notail/noatime opts.
I did not have the fd amounts set, nor ip_local_port_range.
My cache_dirs have each disk mounted as partition, i.e. disk1=/squid1
disk2=/squid2; would your suggestion
be then to halve each disk and partition each as cache_dir? (i.e, go
from squid1-4 to squid1-8 across the 4 disks)
Also have a 5th disk of equal size that has to be used for OS, just fyi
- so these 4 are totally dedicated to Squid.
thanks again, appreciate your input.
-Ryan
