[squid-users] Re: Socket option in Squid

From: Henrik Nordstrom <hno@dont-contact.us>
Date: Fri, 17 Aug 2001 23:49:09 +0200

Squid's network I/O mechanism is in serious need of improvement.
Currently a lot of resources are wasted there. However, as the overall
load increases, the wastefullnes reduces so it is not as bad as it may
seem. In my experiance Squid runs quite well way above where the
CPU gets fully utilized (limit seems to be about 1.5-2 times where the
CPU is fully utilized).

I don't think reducing the amount of reads will make much of a
difference.

There is no "normal" state wrt poll/select. The perhaps best thing to
look at to tell if performance levels are normal is the request latency
measures.

Converting Squid to a forking model is not feasible. If you want such a
design go for Apache. The two I/O models are very different.

--
Henrik Nordstrom
Squid Hacker
Ronald wrote:
> Hello Henrik, I am reading Squid code. My squid 24s1 is becoming
> bottleneck in cpu usage, found that most of the cpu was with poll
> calls. I have read this from one book that we can set SO_SNDLOWAT
> socket option to high value so that we can reduce the number of read
> calls. Does this make sense? And another thing, I have seen in
> cachemgr that no of counters for select is such and such. Is there any
> calculation to check whether Squid is in Normal state like No of
> requests Vs No of poll calls, etc, etc ???? Again, What are the issues
> when we make Squid as a forking a child process per request? Will it
> not be suitable for Squid?  Thanks for any help,Ronald
Received on Fri Aug 17 2001 - 16:01:29 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:01:43 MST