Re: Cache Digests

On Fri, 11 Sep 1998, Niall Doherty wrote:

> Will deletion from Cache Digests be supported in 1.2 ?

Not unless proved to be worth the cost. However, I hope to put some code to
send diffs instead of full digests when possible.
 
> Has the only-if-cached solution to False Hits been implemented yet
> in 1.2 for Cache Digest queries ?

Duane would know for sure. I think we are half-way there. The new
"forwarding" module Duane has been working on will support negative replies
to only-if-cached requests (if it does not already).
 
> Can this be used for HTTP queries that are a result of UDP_HITs ?
> The server side of this should be easy to implement. The client
> side should be the same for a HTTP query resulting from a
> CACHE_DIGEST_HIT.

Yes, I think only-if-cached could be used with HTTP requests after UDP_HITs.
No modifications would be needed on both sides rather than a couple of lines
to add the only-if-cached header for non-parents.
 
> The paper on Cache Digests by DW and AR states that the current
> implementation always forwards directly to the origin server if
> there's no CACHE_DIGEST_HIT. If there are neighbours using ICP
> I presume these will be used. If not, it wouldn't be possible to
> reach the origin server if you're behind a firewall...

This is no longer true. We have a complex waterfall of algorithms from Cache
Digests through NetDB to ICP that are tried in order. The current assumptions
are rather conservative (not trusting Cache Digests 100% when they say
"miss").

> Have there been any other changes in the cache digest implementation
> in Squid since the paper was written for the WWW Caching Conference ?

Yes, and more are on the way. Diffs and separate digests for parents and
non-parents are the major ones, I think.
 
> P.S. Does anyone know if any commercial products are/will be supporting
> Cache Digests soon ?

Not me. I am sure it is just a question of time and adding more cool options
to Squid.

Alex.
Received on Fri Sep 11 1998 - 10:05:32 MDT