W3C home > Mailing lists > Public > ietf-http-wg-old@w3.org > January to April 1997

Re: How to add new "protocols" ?

From: <touch@isi.edu>
Date: Tue, 18 Feb 97 13:01:54 PST
Message-Id: <9702182101.AA01723@oak.isi.edu>
To: luigi@labinfo.iet.unipi.it, jg@zorch.w3.org
Cc: touch@isi.edu, bertold@tohotom.vein.hu, http-wg%cuckoo.hpl.hp.com@hplb.hpl.hp.com, www-talk@www10.www3.org
> Date: Tue, 18 Feb 97 14:19:30 -0500
> 
> As a protocol, HTTP ought to run fine over more or
> less any reliable transport.  It does not depend on any particular
> TCP feature (e.g. urgent data, push bits or delivery of resets).  
> 
> I will note, however, that the ability to close different sides
> of the connection turns out to be an important implementation 
> detail in HTTP/1.1 (see our paper for details), though if this were

I contend that HTTP/1.1 requires more than "any reliable transport",
but more specifically also requires:

	full-duplex
	connection-oriented

Without full-duplex, there is no way to associate a request with
its corresponding response. Even when pipelining requests, 
the assumption is of ordered delivery of requests and responses,
which reliable transactions would not necessarily provide.

Without connections, all of the 'connection' references in the
text must be redefined. I'm not sure how that would change
the protocol, but it seems like an extensive alteration.

Perhaps the above two could be relaxed with
	totally-ordered reliable transactions

but it is not clear at all.


Joe

----------------------------------------------------------------------
Joe Touch - touch@isi.edu		    http://www.isi.edu/~touch/
ISI / Project Leader, ATOMIC-2, LSAM       http://www.isi.edu/atomic2/
USC / Research Assistant Prof.                http://www.isi.edu/lsam/
Received on Tuesday, 18 February 1997 14:11:08 EST

This archive was generated by hypermail pre-2.1.9 : Wednesday, 24 September 2003 06:32:29 EDT