- From: Paul Leach <paulle@microsoft.com>
- Date: Wed, 10 Jul 1996 12:49:36 -0700
- To: 'Shel Kaphan' <sjk@amazon.com>
- Cc: 'http-wg' <http-wg%cuckoo.hpl.hp.com@hplb.hpl.hp.com>
Neat observation. Certainly better than some special purpose hack. But I'll bet not many caches will try the optiimization you suggested to avoid duplicate entries. >---------- >From: Shel Kaphan[SMTP:sjk@amazon.com] >Sent: Wednesday, July 10, 1996 12:19 PM >To: Paul Leach >Cc: 'koen@win.tue.nl'; 'http-wg%cuckoo.hpl.hp.com@hplb.hpl.hp.com' >Subject: RE: Demographics > >Paul Leach writes: > ... > > If fact, "Referer" makes it very hard for a cache to be semantically > > transparent: if a cache was trying to be semantically transparent, the > > presence of "Referer" on a GET request should cause a cache to do a > > conditional GET on the Request-URI. The only way to avoid this would be > > to remember each different value of Referer: had been seen and to report > > hit counts on each of them (somehow). Not only is this more complex than > > using cached 302s, it also would result in exactly as many remembered > > Referers as cached 302s in your proposal. > > > > >Not that I'm recommending this per se, but if a server cared about >this, it could use the Vary header to indicate the result >varies on the Referer request-header, causing any caches to treat the >results independently. > >Of course this would mean a lot more bookkeeping entries in caches, but >implementation-wise, if caches could have unique entries for resources that >were logically different but physically identical (using a digest hash >or something), this wouldn't be that much uglier than some >special case method of keeping track specifically by Referer. > >--Shel >
Received on Wednesday, 10 July 1996 13:02:37 UTC