Re: Cache Digests vs ICP

--MimeMultipartBoundary
Content-Type: TEXT/PLAIN; charset=US-ASCII

On Tue, 28 Apr 1998, Andres Kroonmaa wrote:

> > Not much. ICP, Cache Digests, HTCP, and other peer selection modules are
> > relatively simple and small compared to the rest of Squid. (And you will still
> > need some of them to support a central index server).
>
> Wait a minute, squid as a client of some index server needs only parts that
> query the server and take use of the responses. thats almost all about it and
> almost already in there.
> Having it all inside squid makes it responsible for all the digest creation,
> maintenance of them, expiring, rerequesting, sorting, etc... Is it so trivial?

It is, compared to the rest of Squid. :) This is a minor point though.
 
> ...
> ICP index server of different administartions should peer
> via either ICP or digests, whichever prove more appropriate. In any case,
> any single cache box queries only one (closest) index server and thus needs
> not worry about mixing data from different peers - its a job for index server.
>
> What makes me prefer ICP is the possibility to send ICP messages back to caches,
> like "drop your copy of this url, its old", cache boxes can come and go notifying
> index server about that, and you can have strict mapping between URL and index
> while digests can't. Besides, moving cache-mesh coordination out of squid allows
> to relatively easily change index server algoritms almost on the fly.

Yes, Cache Digests in their current implementation do not support per-request
updates. One of the reasons was that a proxy deletes primarily _old_ objects
from the cache, and nobody requests old objects anyway. Of course, it depends
on many things and is never 100% true.

We are still considering an option of adding "updates" to Cache Digests
(probably as small header fields piggybacked to normal HTTP replies from
peers)..

Alex.

--MimeMultipartBoundary--
Received on Tue Jul 29 2003 - 13:15:48 MDT