Re: [SQU] How to tell if Squid is Overloaded?

>From: Alex Rousskov <rousskov@measurement-factory.com>
>To: Dana Gourley <danagourley@hotmail.com>
>CC: squid-users@ircache.net
>Subject: Re: [SQU] How to tell if Squid is Overloaded?
>Date: Wed, 29 Nov 2000 14:16:43 -0700 (MST)
>
>On Wed, 29 Nov 2000, Dana Gourley wrote:
>
> > I am using Linux-2.2.16
> > I am using currently squid2.4Devel4 with async-io=160, two
> > "cache_dir aufs /cache1/cache 4000 16 256"
> > plus "cache_mem 32MB"
> > I am using two IDE disks. One of them also does the logging.
> > The polygraph test is a modified version of polymix-2. Similar things
>happen
> > with polymix-2. The cache is full for these tests. The peak request
>rate
> > was 74req/sec. Also occurs at 50req/sec
>
>If my math is right, an 8GB cache can hold about 5 hours of fill
>traffic at 74/sec request rate. This may be too small of a cache as
>Polygraph working set is not managed using LRU while Squid cache is.
>
>IIRC, Polygraph reports working set sizes on the console when WS is
>being frozen. See if those values are at least 30% smaller than your
>cache size. Also, keep in mind that PolyMix-3 has a better, more
>realistic working set maintenance algorithms.
>
>IIRC, Squid can report statistics that shows the number of objects not
>cached due to overload. I do not recall the name of the variable
>though...
>
>Alex.
>

The decline in the hit ratio starts in the first hour of a test, so it isn't
a problem of a full the cache (my math agree's with yours).

I found in the cache manager the "Store IO Interface Stats" page which seems
to be what you might be talking about. It had two counters for
"create.select_fail" and "create.success". When I watch these numbers
durring a polygraph test at a high request rate there are alot of
create.select_fails. Can anyone explain what the significance of these
numbers are? "create.select_fail" doesn't sound good what ever it is.

Haven't yet been able to see how these numbers might correlate to a lower
hit ratio. There's an awful lot of create.select_fails right from the
start.

Dana

_____________________________________________________________________________________
Get more from the Web. FREE MSN Explorer download : http://explorer.msn.com

--
To unsubscribe, see http://www.squid-cache.org/mailing-lists.html