From: Marcus Kool<marcus.kool@xxxxxxxxxxxxxxx>
Subject: Re: problem with loaded server
To: "Mustafa Raji"<mustafa.raji@xxxxxxxxx>
Cc: squid-users@xxxxxxxxxxxxxxx
Date: Saturday, April 7, 2012, 10:33 PM
On 04/07/2012 04:37 PM, Mustafa Raji wrote:
hi
i have server which i think it's overloaded
i checked the TIME_WAIT TCP connection status it's
normal and below the limits of the os, and file descriptor
is set for high value and no warning in cache.log about the
file descriptor, so i think it's memory issue or disk
bottleneck, the cache contain 6 hard disk for cache, with 12
cache_dir aufs directory, with 115 GB for each partition my
Ram is 26 GB, calculating that for each 1 cached GB i need
32 MB, so this amount of ram is not enough, of course with
2024 cache_mem, as a solution to the problem i disabled four
partition, and reduce cache_mem to 1024, the server works
find but with low memory in the peak hours it reaches to 400
MB , but no browsing suffering from my client, the problem
could be considered a disk bottleneck and when i disabled 4
partition the problem goes ? of i can consider it as a
memory issue, i realy can't check page fault in the cachemgr
because i don't have the right to install it in working
server, of course additionally
cache.log file show me so many invalid
request, does the number of these invalid request slowing
the performance of the box,
when i disabled my cache directory i put # sign after
my cache_dir directives, does that mean i cant make these
partition back to work, because of fresh object in that
partition not exist, and what squid will do if it caches
objects in the working cache directory and the same object
is exist in the disabled partition, does that effect the
performance of squid if partition enabled a gain, i hope i
was clear in my explanation,
also i have test squid box, i have two squid installed
in the /usr/local/ directory when i start one of them it
does not take the squid.conf that exist in it's etc
directory so, i must start it with the -f option to force it
with the configuration file, what is the cause of these
problem please
thank you with my best regards
http://wiki.squid-cache.org/SquidFaq/SquidMemory:
64-bit Squid needs 14 MB memory index per 1 GB disk cache
Squid has best performance with one cache dir per disk, so
you should reduce the number of cache directories from 12 to
6.
The system needs RAM for OS, file system buffers, other
processes and network buffers.
Let's reserve 2 GB for that.
Memory cache is much faster than disk cache, but the disk
cache is larger.
I suggest to set cache_mem to 8 GB and the total disk cache
to 690 GB (6*115) which requires 690*14=9.6 GB in-memory
index.
Squid will than at least claim 17.6 GB memory. This is
a good starting point for finetuning.
You can probably add more disk cache and/or more memory
cache.
If you run 2 squid proxies, they should have separate disks
for their disk cache and one cache dir per disk.
For 2 equally configured Squid proxies you could use this:
cache_mem 6 GB and total disk cache 345 GB (3*115) which
requires 4.8 GB in-memory index.
Each Squid will than at least claim 10.8 GB memory.
Marcus
