Re: Tuning SQUID

Martin Och wrote:

> If I divide the number of PageFaults - now is the number 143839
> with HTTP requests received: 344239 i get 0,417....
> By http://www.squid.cscnet.cz/Doc/FAQ/FAQ-11.html#ss11.17 this number
> should be between 0.1 - 0.2.

Looking at the pagefaults number alone is not sufficient to determine if
you have a problem, and for most systems it is quite useless as disk I/O
will cause page faults.

The only thing which can be said for sure is that if the pagefaults
number is low then there is no shortage of memory. The rewerse does not
neccesarily apply and depends on a number of things.

For Linux a good tool for determining if you have a memory shortage or
not is to look at the output of "vmstat 5" during peak usage. If you
have more than sporadic activity in the si and so columns then there is
a memory shortage.

If your Squid seems to slow, no activity in the si/so columns of vmstat
and there is more than 50% idle CPU time then enabling asynnc-io may
help you. However, async-io in 2.3.STABLE1 is not as good as 2.2.STABLE5
was.

--
Henrik Nordstrom
Squid hacker