W3C home > Mailing lists > Public > ietf-http-wg-old@w3.org > September to December 1994

Re: Connection Header

From: Henrik Frystyk Nielsen <frystyk@ptsun00.cern.ch>
Date: Sat, 17 Dec 94 16:54:07 +0100
Message-Id: <9412171554.AA10617@ptsun03.cern.ch>
To: fielding@avron.ICS.UCI.EDU, john@math.nwu.edu
Cc: http-wg%cuckoo.hpl.hp.com@hplb.hpl.hp.com

> According to Roy T. Fielding:
> > 
> >    For example, an experimental client may send:
> > 
> >       Connection: keep-alive
> > 
> >    to indicate that it desires to keep the connection open for multiple
> >    requests. The server may then respond with a message containing:
> > 
> >       Connection: keep-alive, timeout=10, maxreq=5
> >    
> >    to indicate that the connection will be kept open for a maximum of 5
> >    requests, but will timeout if the next request is not received within
> >    10 seconds. 
> >
> ...
> > 
> > Comments, please?
> > 
> 
> I have some some serious reservations about this.  Perhaps someone
> more knowledgeable about TCP/IP can alleviate them.  Won't this
> maintain the connection during the entire time that a client takes to
> download and layout and display a document containing containing
> mulitple images?

The idea is that a client can either:

1) Send one request, wait for the response and then send the next one

2) Send multiple responses without waiting for the responses. The
essential thing is that the server always serves the requests in
serial. This is about as far as you can get without a session layer -
Simon ?

Well, let's look at the possible senarios

1) Experimental client / old server: The server will not understand the
header but will serve the first request and close the connection. No
extra round trip time will be needed as the server understands the rest
of the request. It might cause some problems if an eager client sends
more requests to a server without waiting for the response. In this
case some server applications might get confused (even though they
shouldn't)

2) Experimental client / old proxy: This will break if the remote host
does in fact understand the connection header. As Roy points out, this
is a weakness of the design.

3) Experimental client / Experimental server / slow link: The server
will serve the first request but leave the connection open until it has
reached the timeout indicated in the Connction header. New requests
arriving after the closing will be refused. The client should then try
to reopen the connection and possibly use a new Connection header.

4) Experimental client / Experimental server / interrupt: The client
will then close the connection and the server will get a SIG_PIPE as
usual. Any request already in the pipe will get lost.

I still think that it is better to keep the connection open for a small
amount of time as the server socket will go into TIME_WAIT state when
it is closed and then can't be reused before about 240 seconds
(recommended timeout accorsing to Simon)

> By contrast, I would think that an MGET connection sending the same
> data could be a relatively short transaction for the server.  Am I
> wrong about this?  Of course, MGET would require two connections --
> one GET for the HTML doc and image size headers and a second MGET for
> all the images.  But the difference between these two would be
> neglible for the client and potentially a very significant win for the
> server.
> 
> I don't see that MGET poses any problems for proxies.  They would have
> to parse the MGET and decide which files they have cached and which
> they need to fetch.  Old (HTTP1.0) servers and proxies would send an
> unknown method message when they get an MGET and the client would then
> re-issue a sequence of GETs. 

The MGET causes an extra roundtrip time as many servers will think:

	"Gee - nobody has told me about this method. I simply refuse to
	do anything about it"

This is the reason for waiting with a SESSION method to HTTP/1.1 as Roy
points out. Also we would like a more general approach than only having
GET as a possibility. A SESSION method can also be used for interactive
sessions.

However, we need some experimental implementations. The plan is to make
a prototype implementatation of the CERN server and a GUI client (for
example Arena) to see the actual effect when somebody has generated
9Giga of tcp dumps :-)


-- cheers --

Henrik Frystyk
frystyk@W3.org
+ 41 22 767 8265
World-Wide Web Project,
CERN, CH-1211 Geneva 23,
Switzerland
Received on Saturday, 17 December 1994 07:54:41 EST

This archive was generated by hypermail pre-2.1.9 : Wednesday, 24 September 2003 06:31:10 EDT