Re: Squid Weekly Summary Interpretation

On Sat, 24 Jun 2000, Henrik Nordstrom wrote:

> Smells like some kind of log file analyzer to me.
>
> To answer the questions, we need to know what analyzer is being used.

The NLANR access-*.pl Perl scripts by Duane Wessels. Circa 1995 these were
included in the Squid distribution. The Perl scipts are called from the
squid.daily and squid.weekly shell scripts. The latter were undoubtedly
modified by BSDi when they began including Squid as contributed software in
their OS distribution.

> > I have some client objects that have 60% or higher "hit" rates on their
> > request counts. Can an impatient user "artificially" inflate the %hit value
> > by frequently refreshing the display? Or, does this really indicate that
> > the user is frequently accessing a finite set of objects that change
> > infrequently?
>
> Users have a quite hard time to accidently boost the hit rate. If they
> push the reload button then they will force a cache-miss, not a hit..

Only for those objects not in cache. With web pages that exist today, it is
not that unusual to find pages that consist of 20 or more objects. If the
user forces a reload because one gif was not displayed, you might get one
cache-miss while getting 19 cache-hits. Nicht war?

> Having only 60% hit ratio as the highest hit ratio when calculated on
> individual objects are quite low I would say. "hot" objects should have
> a quite high hit ratio, preferably some where around 95% or more.. But
> then I might misunderstand what you mean by 60% or higer "hit" rates..

I would expect to see high hit rates for page objects; i.e. html, gif, jpeg,
etc. I wouldn't expect to see high hit rates for a client object, i.e. my
workstation or an interior Squid proxy cache.

Merton Campbell Crockett
Received on Sun Jun 25 2000 - 11:13:15 MDT