another new paradigm

-----BEGIN PGP SIGNED MESSAGE-----

Here's something I've been playing around with a little...
<URL:http://wwwcache.ja.net/~martin/grand-central.pl>

This is a butchered version of the venerable icpserver.pl which looks
up the requested URL in a DB database, with the intention of returning
an ICP "redirect" to another server. The rationale is that instead of
peering with lots of siblings, you peer with a single machine running
some variation on this code (like a beefy version written in C, or
something :-), and it returns you redirects to the server or servers
which it knows have the URL you asked for.

Now, grand-central could figure out who's likely to have a given URL
if they make it a peer - unless the proxy-only model is being used.
Servers which ask for URLs typically end up caching them later on -
this is the model used by Mirror Image with their Interceptor/
Exchange Point cache architecture.

At the moment I have the building of the database as an separate (and
potentially "offline") process, using URLs harvested from existing
cache servers in our cluster. I've done some small and medium sized
trials, and it looks as though there might be some mileage in this
approach.

The question I've been building up to is - what should grand-central
return if a requested URL is cached by one of the servers it tracks ?
I'm thinking that we could do with a new opcode, which I've been
calling ICP_REDIRECT in my head. What do people reckon to this ?

Ciao!

Martin

-----BEGIN PGP SIGNATURE-----
Version: 2.6.3i
Charset: noconv

iQCVAwUBNTeQeNZdpXZXTSjhAQFFcQQAmQ50esz6cOGlKg0b6OyFHDBI4nArZrcv
cLE52Rk/Ke92BjeNiZUpCgksyROM0V7wvEnt3HeizyTOcN3o1B7UHAqRU6Aauo3u
D/0hsrubZABVLTUehVGwVSDbcfg7G85ky4DlZ2N8Q8iO9kzTgMxo1Hw0xKB4zejY
2dwTjq6OWu8=
=dYw+
-----END PGP SIGNATURE-----
Received on Tue Jul 29 2003 - 13:15:47 MDT