Re: Poor hit/transfer ratio

From: Donovan Baarda <abo@dont-contact.us>
Date: Fri, 5 Dec 1997 23:50:30 +1100 (EST)

Without your proxy many of these would probably result in
complete fetches (since your expire rules will be more generous than many
poorly configured sources).
On Wed, 3 Dec 1997, Ken Piotto wrote:

> HI
>
> Does anyone know if there is a magic number of users (or better, requests
> per unit time) in order to see a respectable hits to kbytes transfered
> ratio?
>
> I'm presently testing squid1.17 on a user group of approx. 20 people within
> my immediate group before releasing to a substantially larger user group
> within our organization. I've been watching the stats for the last few
> weeks and although hit rate seems pretty good (32% of all requests) the
> total data transfer comprising these hits is a paltry 10%. Conversely
> misses is 65% with total comprised data transfer at 89%. Seems the only
> hits occur on relatively small objects.
>
remember that most hits will be IMS type requests, so not the whole page
is transfered, just confirmation that the copy the browser already has is
up-to date. Misses, on the other hand, require a full fetch. Many of
these IMS hits would probably have resulted in full fetches.

What this means is although your 32% hits only comprise 10% of your data
transfer, they have actualy reduced your traffic by more than that.

> Since I'm only seeing about 10000 requests per week from such a small group
> I thought that maybe what I'm seeing is a statistical function of small
> numbers. Still the 32% has me scratching my head. ??????
>
> Oh ya
> squid 1.17, 300M cache, most of which is not used, and 20M Ram.
>
I'm about the same in every respect; users, cache size, ram, and hit
rate. That 32% looks pretty magic allright :-)

> Any words/suggs apprec'd
>

I'm about to experiment with increasing the cache size to 500M. Another
thing that improved performance and received positive feedback from my
users is putting in ACL's to block banner ads. I get about 6% DENIED using
three ACL's, which is more than my hit rate on my parent. That is 6%
blocked that would have normaly been misses :-)

ABO
------------------------------------------------------------------
finger abo@minkirri.apana.org.au for more info, including pgp key
------------------------------------------------------------------
Received on Fri Dec 05 1997 - 04:52:38 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:37:51 MST