- From: Mark Nottingham <mnot@akamai.com>
- Date: Tue, 10 Jul 2001 21:17:38 -0700
- To: Mark Baker <distobj@acm.org>
- Cc: www-ws@w3.org
On Mon, Jul 09, 2001 at 03:58:13PM -0400, Mark Baker wrote: > Mark Nottingham wrote: > > It's not so much that as the decoupling of the request and response. > > In other words, if they make a request, keep the persistent > > connection open, and then send responses upon events, the 1->n > > request/response relationship will confuse proxies. > > Confuse them any more than any other use of persistent connections? Hmm, what do you mean by 'other'? > > Of course, this > > won't be the case if they keep a 1-1 relationship (perhaps by doing > > request...response,request...response,request...). > > They support that mode as well, though as a special case of a > notification replay feature. > > There's pros and cons of both approaches, but at least they provide a > choice. That's very good. > > However, using persistent connections in this manner isn't too > > friendly to proxies; it consumes resources on them, and many will > > close idle connections after a fairly short timeout. When this > > happens, it becomes inefficient to use the HTTP, as you have to > > continually re-establish the connection. > > True, but I believe that's "just" another consideration when deciding > which approach to use. If neither is suitable for you, I suppose the > next step would be to look at something other than HTTP. That has its > costs too. True, as long as their customers go in with their eyes open. Cheers, -- Mark Nottingham, Research Scientist Akamai Technologies (San Mateo, CA USA)
Received on Wednesday, 11 July 2001 00:17:40 UTC