httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Roy T. Fielding" <>
Subject Re: HTTP-NG
Date Mon, 09 Feb 1998 08:28:02 GMT
>> W3C and Xerox PARC's ILU team are working on it with the presupposition
>> that making HTTP look more like CORBA is a good thing.
>  Don't give the false impression that NG will be close to CORBA design,
>we are making progress on the whole NG architecture, but things are not
>stable enough to redistribute too much informations on NG yet. But we
>will definitely try to give more information at the next IETF and WWW7.

Yeah, I was being imprecise.  The presupposition is that what the Web
really needs is an object-oriented API that can use generic middleware
services instead of a standard wire protocol.  It has the advantage that
application creation and evolution becomes easier, but has the disadvantage
that it specifies the implementation instead of the protocol, leading to
lesser performance and failure when the implementation is not sufficient
for some new, unanticipated application, or when the firewall folks
figure out that they are unable to control data transfer.

>  Roy, please forward your comment to the NG mailing-list, we will be happy
>to hear from you. 

I don't like sending comments when I'm not s*bscrib'd.  I'll try to get
myself on the list again.  I have read the entire archive, though I just
went through the latest month tonight.  So far there hasn't been much
point in commenting, since the initial direction of a generic RPC is
completely opposite to what I'd suggest.  It might be more constructive
for me to just go off and define (or help someone else define) a competing
model and then let the two be compared in reality instead of just ideas.

In any case, the HTTP-NG discussion is very informative, and most of
what has been done so far is applicable to both middleware-centric and
protocol-centric models of HTTP.  Likewise, the changes needed for
Apache are the same, except that the middleware-centric approach would
eventually lead to including something like ILU as a server humpback.

>> >Any rough timelines on it?
>> Nope.
>  Yep June 97, at that time we will give an answer to the initial question
>whether an object-oriented model is suitable for HTTP and will allow a better
>evolution of the protocol. Maybe no, maybe yes, in both case we will
>justify our answer with more than words ... 

I was assuming Marc was thinking of a rough timeline for HTTP-NG as
a real protocol.  The June 98 date is just when the initial experiment
is to be finished in order to report some results to the W3C rulers.


View raw message