On 6/11/2011 12:50 a.m., Benjamin wrote:
Hi Amos,
While we are looking for cache_dir consideration , does a single big
directory is better or multiple directory with good amount of size is
better while we are looking for performance ?
Multiple physical disks with cache_dir on each is best. That offers the
most parallal reads/writes. Unless you are adding COSS or Rock storage
cache_dir its not worth placing multiple cache_dor on one HDD. Total
size of any given UFS cache_dir seems not to matter as much as
parallelism on the disk I/O.
And i tried to find from internet that which is better for cache_dir
file system ext3 / ext4 / reseizerfs ( in terms of heavy loaded
systems ) ?
Disabling the journalling actions and atime updating is best for all of
the above. I have not seen any recent speed benchmarks comparing them.
If you base speed on the more generic benchmarks be aware Squid disk
access has a unusually high portion of writes.
Whenever i m trying to find how many squid processes are running i got,
pgrep squid
8311
8313
it means squid always has 2 squid processes. Can we increase it for
high performance.?
One of those will be the master process. One will be the actual worker
process.
You can run as many pairs of mster+worker as you like on a box. Though
we dont recommend running more than one worker per CPU.
See http://wiki.squid-cache.org/MultipleInstances for the relevant
config details for 3.1 and older. If you have squid-3.2 there is a link
there to the SMP support page.
Amos
