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

Re: Sticky stuff.

From: Roy T. Fielding <fielding@liege.ICS.UCI.EDU>
Date: Thu, 08 Aug 1996 17:06:41 -0700
To: Jeffrey Mogul <mogul@pa.dec.com>, Paul Leach <paulle@microsoft.com>
Cc: http-wg%cuckoo.hpl.hp.com@hplb.hpl.hp.com
Message-Id: <9608081706.aa18098@paris.ics.uci.edu>
>     Overall I would like to see an awfull lot of numbers based on 
>     empirical measurement before deciding whether this is a 
>     worthwhile scheme or not. Although it looks OK to me I know
>     from experience that without hard numbers it is very easy to
>     overoptimise corner cases that almost never occur.
>     
> I think it would be nice to get a trace of the actual bytes
> carried by a real proxy (not just the URLs), and apply the
> proposed compression schemes to see how successful they are.
> Of course, this would have to be done carefully to avoid
> breaching the privacy of the proxy's users.

Yes, and I'd also like to see it compared to a complete tokenization
of the protocol (also just a translation table to implement) and the
perceived benefits as compared to a multiplexing of requests on a
single connection. I doubt that there will be any perceptable performance
improvement when using modern user agents (ones without 1k worth of
Accept headers) which do not generate GET requests larger than 2 packets,
since the multi-connection overhead is worse for any congested network.
But, I would like that confirmed (or even better, disproven) by actual
performance data over real TCP connections, and I have no time to do that.

The feedback I received while gathering design issues for HTTP/1.x
is that no piddly performance improvements are worthwhile because
of the added complexity to implementations.  The only changes that
were considered worthwhile were, in order: persistent connections
(mostly to enable continuous connection to an organization's proxy),
multiplexing (mostly to enable non-jerky rendering of the visible
page without opening extra connections), and tokenization of both
header fields and their contents (to reduce bandwidth consumption).

The design of HTTP/1.1 is purposely geared towards tokenization -- that
was the plan all along; once the semantics of HTTP were well-defined, it
would be very easy to make an extremely efficient HTTP/2.0.  Anything
less is, IMHO, a waste of time.  I don't mind if people want to waste
their own time, but I do mind when it is done under the banner of
proposed standard (which ends up wasting everyone's time even when the
idea is a great one). 

I don't see why this can't be done as an experiment first -- I created
the Upgrade header field specifically to allow such experiments to
occur *outside* the standardization arena. Right now we are arguing over
the wording of a proposal which hasn't even been determined to do anything
useful, let alone something that should be standardized.

 ...Roy T. Fielding
    Department of Information & Computer Science    (fielding@ics.uci.edu)
    University of California, Irvine, CA 92697-3425    fax:+1(714)824-4056
    http://www.ics.uci.edu/~fielding/
Received on Thursday, 8 August 1996 17:18:21 EDT

This archive was generated by hypermail pre-2.1.9 : Wednesday, 24 September 2003 06:32:06 EDT