Re: MTU discovery considered harmful?

Philippe Prindeville <philipp@Gipsi.Gipsi.Fr> Mon, 23 April 1990 17:00 UTC

Received: from decwrl.dec.com by acetes.pa.dec.com (5.54.5/4.7.34) id AA00406; Mon, 23 Apr 90 10:00:26 PDT
Received: by decwrl.dec.com; id AA06983; Mon, 23 Apr 90 09:04:43 -0700
Received: from Gipsi.Gipsi.FR by inria.inria.fr (5.61+/89.0.8) via Fnet-EUnet id AA21114; Mon, 23 Apr 90 17:32:31 +0200 (MET)
Received: by Gipsi.Gipsi.Fr (4.12/4.8) id AA20853; Mon, 23 Apr 90 16:37:13 -0100 (MET)
Date: Mon, 23 Apr 90 16:37:13 -0100
From: Philippe Prindeville <philipp@Gipsi.Gipsi.Fr>
Message-Id: <9004231537.AA20853@Gipsi.Gipsi.Fr>
X-Phone: +33 1 30 60 75 25 / +33 1 47 34 42 74
To: mtudwg, smb@ulysses.att.com
Subject: Re: MTU discovery considered harmful?

First, as you point out, your comments are TCP specific.  The
MTU discovery is an IP specific issue.  Clearly, it would be
useful for other protocols (and perhaps TCP as well, I'm not
as yet convinced of the evil you perceive) to have access to
the IP MTU: NFS being the most obvious example.

Now, the question of how TCP uses any information available
to it by the underlying IP is a layering (or more correctly,
service interface) question.  What you seem to have stumbled
upon, and it is no new issue, is that while layering is perhaps
A Good Thing (though Guru Pakular [sp?] might disagree), a
mechanism for exchanging information about desired and avail-
able service parameters is equally A Good Thing.

As a side note, a DS0 is 64kbs (clear channel).  That the
american carriers shave off 1/8 of the bandwidth to be waste-
fully used as signalling (ha!) information is an artifact of
the state of public utilities in the US "free market".

-Philip