[Iccrg] My thoughts on the group, and how to proceed
michael.welzl at uibk.ac.at
Tue Jun 13 11:12:00 BST 2006
First of all, let me apologize for remaining silent for
such a long time. I figure that this is not a good way
to start out as a group co-chair... but we're approaching
the end of the semester here in Innsbruck, with increasing
organizational load. Also, I had some concerns about the
goals of this group, and I needed some time (plus a conf.
call :) ) to get a clear view of the picture.
THE DEPRESSING PAST:
Consider a group of car designers, each standing next to
their cars. All of them have cars which are much better
than the "standard car"; the main company which builds
the standard car has only focused on making it more robust
over the years (hehe, and ensured that it can't overtake
other similar cars (standard-car-friendliness), and that
it will somehow remain useful in case of congestion :-) ).
Now, the way I used to look at this group is like a
collection of these car designers, with the goal of
having them agree on the car of the future (or rather, a
recommendation to the "standard car design company").
I thought that this would be impossible - if you consider
that these people haven't even agreed what metric is
better (maximum speed, better initial speed, won't fall
apart on mountain roads, ...). What's more, I thought that
agreeing on such metrics would also be impossible (I think
we even had a brief discussion in this direction), and
therefore the group could never reach this goal in the
An eventual goal is to produce a recommendation to the IETF on a
solution that would be appropriate for Internet-scale deployment.
(even this additional sentence doesn't seem to help a lot: "It is
possible that more than one solution will be recommended - this is
likely if the solutions can gracefully co-exist.")
THE ENCOURAGING FUTURE:
The way I see it now is different, thanks to Aaron - I'm not
saying that he has exactly my point of view, just that he
made me understand that there *is* a way out of the dilemma.
I say, it's unreasonable for these car designers to try and
agree on a single car, or even a single recommendation.
Instead, they need to step back and see what they learned
from what they built. For instance, if 5 out of 10 car
designers achieved higher speeds by building the motor in a
special way (this is the part where you learn that I know
absolutely *nothing* about cars :-) ), it might be possible
that all car designers agree that this way of building a motor
is the right thing to do.
Let me make things a little more concrete at this point:
both CUBIC and HighSpeed TCP, for instance, increase the
sender rate faster than TCP *IFF* there is little loss in
the network. Based on experience with these protocols and
perhaps others which are similar in this aspect, we could
agree that this is the right thing to do and turn it into
a general rule - a recommendation for future congestion
These are rules that we could derive by learning from
performance enhancing aspects of the various congestion
control mechanisms out there. I think that we shouldn't
even try to work towards a concrete recommendation that
says "this is what your future cc. mechanism should look
like", but really have broader rules that can serve as
a framework for future designs.
More information about the Iccrg