RE: Suggestion for NEW Issue: Pipelining problems

I think Yngve is primarily pointing out that a significant body of the deployed base of HTTP servers do not support pipelining.  Unfortunately, there's no reliable method, a priori, to determine whether or not the remote server is going to handle pipelining correctly, or whether it will either crash or deliver incorrect results.

I don't know that ambiguities or inaccuracies in the RFC are to blame-- I think the problem is that most servers never bothered to try to handle pipelining correctly.  Today, there is a chicken and egg problem: without broader client support for pipelining, servers have little real incentive to do the work to support pipelining, but the large number of broken servers means that the cost to implement pipelining in the client skyrockets.  As the only popular browser that implements and enables pipelining by default, Opera is leading the way from the browser space.

Proposed modifications I've seen to help reduce the need for client heuristics include:

        1. The server provides some form of indication that it intends to support pipelining
        2. Extending #1, the server's response in some way indicates which request it matches (e.g. client request header and matching server response header)

I'm sure there are many more approaches.

Eric Lawrence
Program Manager
Internet Explorer
Want to view and tamper with HTTP(S) traffic?  Try Microsoft Fiddler
Install IE7 - http://install7.com


-----Original Message-----
From: ietf-http-wg-request@w3.org [mailto:ietf-http-wg-request@w3.org] On Behalf Of Travis Snoozy
Sent: Tuesday, July 17, 2007 7:22 AM
To: Yngve N. Pettersen (Developer Opera Software ASA)
Cc: ietf-http-wg@w3.org
Subject: Re: Suggestion for NEW Issue: Pipelining problems


On Tue, 17 Jul 2007 13:04:14 +0200, "Yngve N. Pettersen (Developer
Opera Software ASA)" <yngve@opera.com> wrote:

>
> Hello all,
>
> I thought I should mention this particular issue I am seeing.
>
> There are a number of servers that does not handle pipelining at all
> well.
<snip>

... and?

More important than pointing out a bunch of bad implementations is
to point out what's wrong with the RFC that led to the bad
implementations. However, pipelining is pretty cut-and-dried -- I don't
know that the RFC can make it much clearer than it already is. If you
see ambiguities in the RFC that lead to these differing server
behaviors, though, please do identify them.


--
Travis

Received on Tuesday, 17 July 2007 15:33:18 UTC