- From: Scott Lawrence <lawrence@agranat.com>
- Date: Mon, 31 Aug 1998 22:25:39 +0000
- To: HTTP-NG Comments <www-http-ng-comments@w3.org>
- CC: iesg@ietf.org
The HTTP-NG BOF in Chicago was very helpfull to me both in understanding what has been done in the W3C Activity and in clarifying my thinking about how it should proceed within the IETF. My view of the essential elements of what HTTP-NG has done so far is that they have described: 1) A framework that more precisely specifies application message semantics (the object-oriented NG interfaces). 2) A more compact binary encoding for Web application messages. 3) An application message boundary mechanism (the webmux fragments). 4) An application association mechanism that supports multiple interleaved associations over a single transport layer connection (webmux sessions), including: 4a) the association setup uses a very simple endpoint definition is be easily inspected by firewalls. 5) A mechanism for describing a service stack below the application to provide additional services (compression, security). Despite the fact that many other proposed protocols have been choosing HTTP as a 'transport', I was suprised to hear that the HTTP-NG effort was being considered within the Transport Area rather than Application. The main arguments that I heard in Chicago in favor of this were: that other protocols were doing it anyway (people are making a poor choice for the wrong reasons, so we should make a poor choice that will validate it?); and that webmux credits would have dangerous potential interactions with TCP flow control that needed transport expertise to understand (very good reason!). Having spent some time thinking about the list above, I have come to believe that some of the reasons why so many recent applications have been choosing HTTP as a substrate (it isn't a transport no matter how many times you call it one) is that it provides (3), (4), (4a), and (5) from that list. These things are important requirements of a very large subset of all application layer protocols, and TCP does not provide them. I don't believe that a case can be made for putting work on (1) or (2) into the Transport Area, and I'm not even convinced that an object-oriented (or other) interface definition framework is a good direction for HTTP to go. A binary encoding for HTTP/1.1 would be easy to do without the new paradigm, and would be a very interesting experiment. I do believe that 3 through 5 are desperately needed, and while my computer science professors would all have called those Session Layer, I'm willing to call them Transport if it means we can get them done once and well so that other applications can use them. With that in mind, I humbly offer my suggestion for a Working Group name: Enhanced Application Transport Services (eats). -- Scott Lawrence Consulting Engineer <lawrence@agranat.com> Agranat Systems, Inc. Embedded Web Technology http://www.agranat.com/
Received on Monday, 31 August 1998 18:25:18 UTC