- From: Marc Slemko <marcs@znep.com>
- Date: Wed, 11 Mar 1998 23:23:18 -0700 (MST)
- To: Dan Connolly <connolly@w3.org>
- cc: www-talk@w3.org, frystyk@w3.org, jg@w3.org
On Wed, 11 Mar 1998, Dan Connolly wrote: > Marc Slemko wrote: > > There are plans for things like HTTP/1.2 (relatively minor) and HTTP-NG > > work is underway that addresses a lot more. But you can't find out > > anything about HTTP-NG except for the very limited info available at W3C > > unless you are special. While you may be, I'm not. > > It's important to me that the community believes W3C is > running HTTP-NG in a reasonable way. I infer from your > comments that you think we're not. So I'll take the opportunity > to try to convince you. I don't think that the development of HTTP-NG is not being done in a reasonable way (note the double negative); I can't think that because I have nothing to base such a viewpoint on. In the absence of more information, and knowing who is involved, I can only guess that it is more or less reasonable. I do have concerns about the interaction between this process and the rest of the world. There is almost no information available at http://www.w3.org/Protocols/HTTP-NG/ regarding it, and it is not clear if any of the information that is there is valid or is just leftover from some ancient history (eg. the other month, the IETF working group site was suggesting 1.2 would be submitted to the IEST in October 96). (well, the information there does appear to have been improved a bit since I last looked a month or so ago.) Sometimes, protocols get stalled and never end up working out properly. eg. SNMPv2. What I have heard about an object oriented architecture for HTTP-NG makes me concerned that the same thing is possible. From my network perspective, that is bad. There are serious issues with HTTP/1.1 <-> network interactions, etc. that need to be dealt with in a reasonable timeframe. I went through this whole thing with Henrik a month or so ago and will not go into the full details again because there is no point. He was upset that a total of around a dozen lines of information saying that yes, HTTP-NG did exist and was being worked on, and a very short outline of the overall concept, were posted to the Apache development mailing list. Let me be very clear that I fully understand the issues involved with working on such a project and that a closed development environment may well be the best solution. There may also well not be anything to make public yet. But I don't know that. I am not suggesting that all development should be open or even that there is any detailed information to be made available at this piont. Henrik's guideline is that someone should be willing to spend 50% or more of their time on HTTP-NG work to be considered. That effectively completely cuts out anyone deeply involved with current real-world webservers, and anyone not part of a company willing to fund the development of HTTP-NG. I am concerned about what sort of direction that could lead to. By the time a draft is released, it is far too late to go changing basic design issues that may be broken. No insult intended to the people that are involved and I have great respect for their knowledge and abilities, but I am concerned about the lack of room for early feedback and viewpoints both from current HTTP implementors or to them. I am not up to the design of HTTP-NG, but I would like to think I have a little experience with HTTP and a little idea of what my concerns are. He also mentioned some mytical "HTTP-NG Interest Group" that others could participate in, but I could never get any more information about it. For example, it is completely impractical for the current Apache server to implement multiplexing. We are in the process of designing a framework for 2.0 that will change this and probably perhaps maybe (since I can't even guess at what HTTP-NG will do) make HTTP-NG drop in, but if we weren't it would certainly lead to a very large lag time before Apache would ever properly implement HTTP-NG at more than a trivial level. The complete lack of information on nearly everything related to HTTP-NG means that a complete redesign or extensive changes may be required to implement it after the Apache 2.0 framework is established. That would be unfortunate since it certainly would delay Apache's implementation. While I suspect it is nonsense to be talking implementation at this time, when looking at designs for the Apache design, and no one knows if HTTP-NG will actually make it that far, I don't care about the protocol details just the concepts. For example, I'm trying to figure out the issue of just how you plan to allow for efficient server-side implementations of multiplexing without embedding SMUX (or whatever is used) in the kernel. Apache 2.0 will improve performance with numerous optimizations using more advanced IO models; but then when you add in multiplexing you throw many of them out the window again. To summarize, I have nothing to complain about regarding the direction the HTTP-NG protocol is going and that is the problem. You can either take it from the perspective that I just like complaining or that I have concerns about the future of HTTP and the current information available on HTTP-NG does nothing to allay them. -- Marc Slemko | Apache Group member marcs@znep.com | marc@apache.org
Received on Thursday, 12 March 1998 01:28:20 UTC