------- start of forwarded message (RFC 934 encapsulation) -------
Sender: owner-end2end-interest@ISI.EDU
To: sthaug@nethelp.no
Cc: end2end-interest@ISI.EDU
Subject: Re: Performance vs Bandwidth Paradox
Date: Wed, 30 Sep 1998 11:20:06 -0700
*> I don't see that this has anything to do with end2end-interest, though.
*>
*> Steinar Haug, Nethelp consulting, sthaug@nethelp.no
*>
Yes, thank you. The intended focus of this mailing list is research
issues in end-to-end protocols and services.
Van Jacobson and others gave us a TCP that is superbly engineered to
adapt very well over a very wide dynamic range -- as long as the TCP at
the other end of the wire is another VJ TCP. Although it is adaptive,
a VJ TCP still has built in a number of "magic numbers", like the
initial RTO, the min RTO (timer granularity), and the EWA constants.
Van has argued that his particular choices of magic numbers are in some
sense optimal. His arguments sound convincing at the moment, but I am
always left with a queasy feeling... As the earlier comments about
Solaris TCP vs. BSD TCP illustrated, small and apparently reasonable
changes in these magic numbers can cause bad things to happen, like
useless retransmissions.
Can we create a reliable transport protocol that is adaptive on
all dimensions, i.e., has NO magic numbers?
Bob Braden
------- end -------