Re: Cisco Cache Director (Was RE: Does Squid beat the rest? )

Hi

> There is at least one interesting thing in the Cisco setup, which is *possibly* translatable into Squid.

It is

> On the other hand, the Cisco router computes a hash from the HTTP server name and uses this hash to direct to one cache of the farm. Thus, requests to one given HTTP server will always end to the same cache. No duplication! The router just has to know the IP addresses of the members of the farm.

Yep - this is a really great idea, and is used in the latest Microsoft proxy
as far as I know.... have a look at
ftp://ftp.is.co.za/internet-drafts/draft-vinod-carp-v1-00.txt

Basically:

You can change your java auto-config file so that it creates a hash using
the URL that depends on the number of caches. For each URL it then connects
to the appropriate cache... which means that you will probably get a HIT
straight off.

squid can't talk to a parent like this though...

> Does it sound like a good idea?
Yes... if all of your clients run the auto-config... if not (like us)
there won't be any major advantage until squid can decide which siblings
to query depending on the URL. (if you run a cache-conglomerate like us)

Oskar
Received on Mon Sep 29 1997 - 08:45:45 MDT