Re: HTTP: multiple-get (was: T-T-T-Talking about MIME Generation)

On Thu, 15 Dec 1994, Daniel W. Connolly wrote:

> Ah... that reminds me: before we forget completely about HTTP/1.0, I
> think the "Pragma: keep-connection" proposal (the idea; not
> necessarily the syntax) is worth persuing in the short term.
> 
> In other words: I agree that we should enhance clients and servers to
> conduction multiple HTTP/1.0 transactions over the same connection.
> 
> The Content-Type issues are a little sticky, but it's probably worth
> persuing...

It was pointed out at the IETF HTTP-ng BOF that the keep session open
approach (whatever the syntax) applied to the client/proxy and 
proxy/server connections independently.  As such the client/proxy could
end up open a long time.  (I would suspect a high performance proxy 
server could to virtual switching between client connections and
idle server connections.) As I understood the discussion, multiple
complete HTTP transactions would travel down the pipe so Content-Type
issues should not vary from the current connection/transaction model.

I think nothing I heard precludes a client from holding sessions open
with multiple hosts or multiple sessions with one host.  I for one
find the human factors of the NETSCAPE approach much more satisfactory
with early rendering of available information then sitting idle for
long intervals waiting for a stupid company logo to be transfered or
what ever.  The SGML organization home page is particulariliy offensive
as the first graphic fills the screen.  Infering from observed 
behavior, I would speculate that the primary reason NETSCAPE opens
multiple connections is to learn the shape of graphics to be included
thereby making the rendering more useful.

It might be nice in a future HTTP to consider restartable transactions
such that a user interrupted transfer could be restarted at an 
intermediate point ... for that matter a long file transfer could
restart.

Dave

Received on Thursday, 15 December 1994 22:56:19 UTC