- From: N.G.Smith <ngs@sesame.hensa.ac.uk>
- Date: Thu, 11 Apr 1996 09:22:33 +0000
- To: Luigi Rizzo <luigi@labinfo.iet.unipi.it>
- Cc: http-caching@pa.dec.com
>Perhaps a bit off-topic, but perhaps I have seen (and now lost >reference to) some posting related to this on the list. > >Some caching protocols try to optimize the use of memory by >determining which documents are more "popular" and pulling them >towards clients, or pushing them away from servers. The popularity >of a document is often if not always based on incoming requests. > >One difficulty with this approach is that once a document has been >replicated elsewhere, and caching is effective, requests to the >original source (be it the server or a proxy) are greatly reduced >and this disturbs the protocol. > >Is there a way for a cache to provide feedback to the original >source reporting access statistics for a given document ? Honestly >I am not even sure if this is something that belongs to HTTP or >not. On one side, traffic related to cache-management (not directly >generated by clients' requests) probably should travel separately; >on the other side, clients' requests may have so many side effects >on caches that it might be easier to mix the two things. > > Luigi This was raised as a very important issue at the W3C Demographics workshop back in January. As a result the www-logging@w3.org mailing list was created. A couple of working drafts have been written but otherwise things seem to have stalled. This work is crucial to caching. Already we are seeing plummetting hit-rates as advertising funded services make their pages uncachable. I know that there has been some discussion on this list, but the possibly divergent solutions need to be tied together. The `token' system is surely the more bandwidth efficient, but the feeling coming from the advertising people is `we want the log of every transaction' - why? - 'Because it's possible'. Neil.
Received on Thursday, 11 April 1996 09:54:32 UTC