Re: Poor hit/transfer ratio

From: Gero J. Dittmer <gdittmer@dont-contact.us>
Date: Thu, 4 Dec 97 09:09:45 +0100

>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.

This question is easily answered from your logs. Most of your hits will
be IMS_HITs which means that the browser has the information requested in
cache and does only an IMS request to make sure it is still current. That
can be answered by the cache in a few hundred bytes and some
milliseconds. This behaviour is common to all kind of caches (not just
web caches). Their design goal is speed and they save you more latency
than data volume.

The most impressive number about a squid cache is not the percentage of
bytes saved but the transaction speed measured in (bytes transferred /
elapsed time). My users receive 50% better performance than comparable
companies. Another effect is buffering. A Mac / W95 machine has not true
multitasking and might not always be ready to receive incoming data fast
enough. Squid does this in behalf of the client and forwards as fast as
the client can receive with no delay to the original transmission from
the web.

Regards, Gero

-----------------------------------------------------------
--- Ammirati Puris Lintas ---
--- Gero Dittmer, Mittelweg 177, 21048 Hamburg, Germany ---
--- phone: +49 40 41441217 fax: +49 40 41441616 ---
--- Internet: gdittmer@aplintas.de CIS: Gero_Dittmer ---
--- ... waiting for a unicorn under chapter eleven ... ---
Received on Thu Dec 04 1997 - 00:14:43 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:37:50 MST