W3C home > Mailing lists > Public > ietf-http-wg-old@w3.org > May to August 1996

Re: Sitcky headers and pipelining (was: Sticky header draft -- as an attachment)

From: Jeffrey Mogul <mogul@pa.dec.com>
Date: Wed, 07 Aug 96 13:40:09 MDT
Message-Id: <9608072040.AA01162@acetes.pa.dec.com>
To: Paul Leach <paulle@microsoft.com>
Cc: http-wg%cuckoo.hpl.hp.com@hplb.hpl.hp.com
X-Mailing-List: <http-wg@cuckoo.hpl.hp.com> archive/latest/1227
    >From: 	David W. Morris[SMTP:dwm@shell.portal.com]
    >On Mon, 5 Aug 1996, Jeffrey Mogul wrote:
    >> I think this leads to an ambiguous situation when the client is
    >> pipelining requests.  We identified this ambiguity at the meeting
    >> we had in January of the persistent-connections subgroup.
    >I think the ambiguity is also resolved if we change HTTP/1.1 to not
    >allow pipelining UNTIL the first server response is received accepting
    >the persistent connection and hence pipelining.
    I thought that this was already the rule. Did it get lost when we made
    persistence the default?

Not lost, exactly.  My recollection of the situation is this:  The
draft-*-06 spec does not require the client to wait before pipelining.
Dave Morris asserted that this could cause problems; Roy Fielding
asserted that it would not.  Larry Masinter asked some people to
consider the issue.

My position was that I was not convinced by either side.  I suggested
that if Dave (or someone else) could come up with a specific example of
how this would break, then I would be more sympathetic to his point of
view.  As far as I remember, nobody has provided such an example.
Given that the "Proposed Standard" phase is the place to take this kind
of risk, I felt it was not necessary to require the client to wait.

Note that if waiting is actually unnecessary and yet the spec requires
the client to wait, we are wasting an entire RTT on many (although not
all) HTTP connections.  I suggest that this would be a mistake.

In particular, if you agree with me that the existing sticky-header
proposal has the race condition that I believe it has, it would be
possible to eliminate it either by requiring a client to wait before
pipelining, or to use the three-way handshake that I described.  But
because of the extra RTT delay the former option imposes on pipelining,
it seems preferrable to avoid it if possible.

Received on Wednesday, 7 August 1996 13:49:55 UTC

This archive was generated by hypermail 2.3.1 : Wednesday, 7 January 2015 14:40:17 UTC