[Iccrg] Heresy recapped
mathis at psc.edu
Mon Apr 14 05:16:38 BST 2008
My "heresy" proved to be rather anticlimactic. Although there was a spirited
discussion about the nature of fairness*, there was no real discussion about
my position about rescinding the TCP-friendly paradigm. Does this mean there
is agreement in this group?
I would like to nudge the iccrg to go on to the next step: What do we need to
do to phase out the TCP-friendly paradigm?
I had a separate conversation with the original authors of "TCP-Friendly",
Jamshid Mahadavi and Sally Floyd , where we came to realize that there
still needs to be some litmus test for acceptable protocol behavior.
There is a huge amount of work needed here: we do not even have a formal
definition of congestion collapse, much less any methods to test if a given
congestion control algorithm might be subject to collapse. Is it sufficient
for all users/protocols to merely optimize their own good put? There is an
entire arena of research here that has been waved away by "its TCP-friendly,
so we don't need to worry about ...." All past questions potentially need to
In the short term, we can always evaluate protocols and algorithms on a
case-by-case basis. In doing so we are likely to discover the underlying
principles. I guess the very short term question is what, if anything, should
we say to tcpm about BIC and CTCP?
Note we probably need a warning labels on some protocols: "This protocol (or
application) will not politely share the network with other users, unless
network itself enforces appropriate sharing. In environments without the
required controls in the network it is likely to cause lockout or other
*) I specifically did not define fairness because I believe that my comments
apply independent of the fairness model. Fairness is very important to ISPs,
and their conversations with both their customers and their equipment
suppliers. Except for algorithms where people want to allocate header bits or
code points, I don't believe that these conversation need the participation of
standards organizations because reasonable differences in implementation do
not have any affect on interoperability. Furthermore, it surprises me that
some of the ideas are not held more closely, because this is an ideal arena
for ISPs to have proprietary solutions that are better than their competitors.
 Jamshid Mahdavi, Sally Floyd, "TCP-Friendly Unicast Rate-Based Flow
Control" http://www.psc.edu/networking/papers/tcp_friendly.html, January,
Thanks for the feedback!
Matt Mathis http://staff.psc.edu/mathis
Evil is defined by mortals who think they know
"The Truth" and use force to apply it to others.
More information about the Iccrg