Re: Paying for someone else's traffic?

On Sun, 1 Mar 1998, Dancer wrote:

> Hmm. I seem to remember that "the ICP query does not contain any HTTP request
> headers which may affect the reply."

Damn, that's right. :-/

> Since the ICP request carries nothing about this, other than the URL
> (maybe it should?)...

Maybe it's time to start work on the next revision of the ICP protocol
spec? This problem appeared when "miss_access" first came on the scene,
and will no doubt appear in some way again in future until the underlying
protocol itself is "fixed".

> When it's requested, however, it's ability to protest that "You didn't
> tell me you wanted a _new_ one..." is only limited. It admitted to
> having it, but we said the age didn't suit us.

Maybe .. just maybe; how about this - when you receive a request via HTTP
that includes a "refresh" component, check the "miss_access" rules .. if
the object is in the cache but the client is requesting a refresh *and*
according to the "miss_access" ACL that client shouldn't be allowed to
request objects that are not already in the cache, then *ignore* the
refresh and just return the object in the cache.

How's about that? Gives protection to the proxy answering the query, yet
doesn't cause any bogus remote proxy error to be returned to the end-user.

> I'd recommend either having such conditional refreshes _never_ go to
> neighbours, or figure out some way to communicate desirable age-data.

Until ICP can be "fixed", perhaps the former is the way to go (coupled
with the "protection" afforded above).

Cheers..

dave
Received on Sat Feb 28 1998 - 20:04:07 MST