- From: Poul-Henning Kamp <phk@phk.freebsd.dk>
- Date: Sat, 31 Mar 2012 12:13:38 +0000
- To: Mark Nottingham <mnot@mnot.net>
- cc: Willy Tarreau <w@1wt.eu>, "<ietf-http-wg@w3.org> Group" <ietf-http-wg@w3.org>
In message <A90E6D81-FA41-4316-8650-5CB16158AE76@mnot.net>, Mark Nottingham wri tes: >We can discuss the problem of date generation/parsing. In a 2.0-only >chain, it would indeed be nice if we could dispense with this altogether >(e.g., with a separate set of headers to replace >date/last-modified/expires that get transformed to them on a 1.x hop >only). Let's discuss that. Yes, we absolutely need to do that. But there are things we should think about before that, such as "Why do we even need a Date: header in the first place" ? It's not like we're talking store-and-forward messaging... My preference for how to shape the WG progress, would be to set up a number of straw-man scenarios, and work through them to understand where it is HTTP/1.1 come up short, and what HTTP/2.0 needs to do better. For example, if one of the strawmen is "1 Tb/s HTTP served on a single IP#" we will probably end up with an analysis like: The first thing our incoming requests sees is a load-balancer/director gadget, which would much more precisely be named a "HTTP-router" Typically requests are distributed to backend HTTP servers based on relatively crude metrics, typically: Host: URI suffix (".jpg" vs. ".html") Session(-cookies) Extrinsic information (server load, etc.) While compression in general is a desirable thing, there may be a compelling argument for at least making it possible to avoid/prevent compression of these "envelope-fields" in order to reduce the amount of work a http-router has to do. The parallels to flow-routing should certainly be explored, in particular with respect to a session-concept that doesn't rely on cookies. Ohh, so having a robust session-mechanism is important at the second-lowest level of HTTP transport ? Hmm, maybe we should pay some attention to that, rather than have people kludge it with cookies. And so on... HTTP/2.0 needs to be a visionary standard, otherwise we'll just be slowing down progress. -- Poul-Henning Kamp | UNIX since Zilog Zeus 3.20 phk@FreeBSD.ORG | TCP/IP since RFC 956 FreeBSD committer | BSD since 4.3-tahoe Never attribute to malice what can adequately be explained by incompetence.
Received on Saturday, 31 March 2012 12:14:03 UTC