> I thought we would back off to a smaller packet size if we "lost
> contact" after N seconds.  Am I wrong?

I think that is fine, as long as N is in terms of RTOs.  In other
words, the sending rate should not be greater than allowed by the
TCP congestion control algorithms.

Kevin Lahey has documented some of the problems TCP implementations
have with respect to pmtud in the following Internet-Draft...

    Kevin Lahey.  TCP Problems with Path MTU Discovery.  August,
    1999.  Internet-Draft draft-ietf-tcpimpl-pmtud-02.txt (work in
    progress).

The document is currently under discussion in the IETF's tcpimpl
working group.  Comments/discussion about the draft should take
place on the tcp-impl mailing list (information about the list can
be found at:
http://www.ietf.org/html.charters/tcpimpl-charter.html).

allman


---
http://roland.grc.nasa.gov/~mallman/