Date: Sun, 8 Feb 1998 21:57:52 -0700 (MST)
From: Marc Slemko <marcs@worldgate.com>
To: new-httpd@apache.org
Subject: Re: HTTP-NG 
In-Reply-To: <199802090436.XAA26283@rufus.w3.org>
Message-ID: <Pine.BSF.3.95q.980208214639.8454T-100000@valis.worldgate.com>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Sender: new-httpd-owner@apache.org
Precedence: bulk
Reply-To: new-httpd@apache.org

On Sun, 8 Feb 1998, Daniel Veillard wrote:

> > >Any comments on what all will be included that could require consideration
> > >for 2.0?
> > 
> > A protocol-independent core and multithreaded abstraction and stream I/O.
> > In other words, what we knew would be needed all along.
> 
>   The goal is to be able to run HTTP apps on top of HTTP-NG so we shouldn't
> need absolute change in the API to have NG into Apache. Just keep a good
> layered API for all the protocol stuff, avoid mixing the calls to the
> network API and raw write to streams, and it should be Ok.

One of the features is to allow multiplexed streams, right?  Nothing
useful on the w3 site on it, but it appears that is still the plan.

To do this with good performance requires a lot more out of a web
server than serialized requests do, even though the overall
workload may be lower due to a lower number of TCP connections.

Essentially, you have to fully move away from a process or thread
based model and go to something like Dean's process model; at the
very least, you need the ability to have a pool of threads where each
request will be serviced by a thread then the association will be
severed until work is needed again, at which time a different thread
may pick it up.  Things like async writes (ie. aio_*, completion
ports) fit into this.

There will be major changes in 2.0 anyway, and knowing what sorts
of things HTTP-NG will be needing for best implementation is useful.