On 12/05/2013 6:51 p.m., Ahmad wrote:
hi , i have squid server that can make about 70 M in the rush hour , i doubt that there is limitation and i can make better than 70 M be if i changed ulimits value .
I assume you mean 70M Bps and not 70M Rps (which would be a really wonderful sight to see for Squid).
The current squid releases are known to service somewhere between 50-100 MBps. So your 70M is about right for a Squid reaching its peak performance capacity. The exact value depends on traffic message rate and whet the proxy is configured to do. So please take this with a grain of salt and consider that it may be a problem or your proxy might just need expanding.
i want to ask , if i type ulimit -n its only 1024
<snip>
i restarted my server , and still ulimit -n give me 1024 not 65536 !!! ================================================
That does sound bad, BUT. Be aware that the limit you get told there is just teh default, it is not necessarily what Squid is running with...
im using debian os : root@vyatta:~# cat /etc/debian_version 6.0.5 =========================
The Debian package init script alters limits for Squid. Please check your /etc/init.d/squid script to see what ulimit is setting there. Alternatively you can check your cache.log messages for Squid reporting how many FD it is running with. This may be in /var/log/messages instead of cache.log.
in my squid.conf : =============================== max_filedescriptors 65536 max_open_disk_fds 65536 ========================
This is not great. Squid uses up to 5 FD per client connection, only one of which is a disk FD. Your traffic rate indicates this is not auseing you problems. But I still recommend tweaking those values so disk FDs was a bit lower than max FDs by at least 10%.
my question is how to raise this value to 65536 ???
max_filedescriptors tries to do it. The Debian init script will be using ulimit to raise it to something as well.
does raising this value make a big performance to squid ??
"It depends". If you are really running with 1024 FD and getting 70MBps it could go either up OR down. What it will do is allow more client connections through - which will expand the traffic capacity for more parallel requests. But that might also mean concurrent requests slow down as Squid is handling more in parallel.
If your Squid is already running with 64K FDs then the bottleneck issue is something else entirely.
Amos
