- From: Shel Kaphan <sjk@amazon.com>
- Date: Mon, 12 Aug 1996 11:26:17 -0700 (PDT)
- To: jg@zorch.w3.org
- Cc: hallam@etna.ai.mit.edu, http-wg%cuckoo.hpl.hp.com@hplb.hpl.hp.com
jg@zorch.w3.org writes: > If a user agent profile is a URL, rather than some silly static > database, there is no monopoly involved; a server would just > fetch the profile from the URL when first talking to the client, > and cache it. > > Unless such a cache of profiles were tiny, it is very likely that > any significant server will have profiles for almost any user agent > in use (and you can roll your own, anyone who has the ability to > provide a web page, and since this seems to be standard these days > with Internet accounts by ISP's, this means everyone). > > This is the scheme Simon Spero suggests in his NG work, and it looks > like a fine one to me. > - Jim > > I definitely like the level of indirection implied by using a URL for this, but I don't like the reliance it puts on a specific "profile server" being available. It doesn't sounds like a solution that scales all that well. Let's not underestimate the number of servers that may exist in the future. Can we "prove" that the number of servers and number of different UAs is sufficiently small to allow this kind of scheme to work? Instead of using a URL, could we not arrange to use the USER_AGENT string as a key to databases maintained by replicated servers, using some standardized URL-based addressing scheme to fetch the profile document associated with a specific user agent? Then it would be a matter of server configuration to tell it where to look for this database. It has a bit of the flavor of DNS. --Shel
Received on Monday, 12 August 1996 11:47:55 UTC