Keith,
At 2:56 pm +0100 23/11/98, Keith Moore wrote:
>It would help. But basically I want to see an architecture which
>supports all of:
>
>direct client access to supplier-provided replicas
> (on referral to those replicas via a URI resolution mechanism)
>client access via intermediary to supplier-provided replicas
>client access via intermediary to cached replicas
As I suspected, the problem (for me anyawy) seems to be one of definitions.
I was defining an "intermediary" as any entity which holds a copy of an
original resource. There are, of course, architectural differences between
a (traditional) "proxy cache" and the sort of replication system which
would be envisioned in a scheme where URI resolution was widely deployed.
In particular, the both the user's and publisher's interaction with both
systems may be quite different, not to mention the replication method
itself.
So, in short, I agree with all your scenarios - and there are probably more
besides.
>I don't agree with the assumption that an intermediary should be
>required. Intermediaries significantly decrease correctness,
>transparency, and reliability, and add a huge amount of complexity,
>for what may well turn out to be, in the long term, minimal gain.
>I want an architecture that can do without intermediaries, or
>minimize their use, when it makes sense to do so.
Well, I dont think it is an assumption but rather a statement of the facts
as they currently stand. Part of the charter of this working group proposes
to document current practice and current practice is that there are many,
often non-interoperable, caching solutions out there which would benefit
both from documentation and standardization work. So, if you can accept
that we are talking about intermediaries as referring to *all* entities
holding replicated copies of an original, then this fits well into your
above scenario.
I have updated the charter already today but will try to add this too, if
you agree.
Regards,
John
--
"The Challenge of Gigabit Networking"
The TERENA-NORDUnet Networking Conference 1999
7-10 June 1999 Lund, Sweden
This archive was generated by hypermail 2b29 : Thu Nov 18 2004 - 11:21:25 MST