[e2e] TCP in outer space
J. Noel Chiappa
jnc at ginger.lcs.mit.edu
Sun Apr 15 12:53:29 PDT 2001
> From: Alhussein Abouzeid <hussein at ee.washington.edu>
> If I may dare point out without the risk of being personally
> attacked .. and hence being considered ignorant, traitor or
> anything of the like
I don't have any problem with people having radical ideas. (I've had a
few myself, in fact!:-) In fact, I'm rather partial to them, because
that's the way you make big jumps.
I just generally find them more useful when they are based on a full and
complete knowledge of the field and its history.
>> The Internet's congestion control mechanisms actually work pretty
>> well .. - in part because of clever design which informs the
>> characteristic that if people try and get more than their "fair
>> share", they almost always actually get less.
> the statement .. (the "fair share" claim) is not correct. Using the
> same loose language, it is well known that there are many
> techniques that "people" can use to get more than their "fair
> share" [of a link capacity] (e.g. multiple concurrent TCP flows,
> non-responsive flows, large initial windows, etc.).
My comment was a (simplistic, to keep it short) rephrasing of something I
seem to recall Van Jacobson (at least, I'm pretty sure it was Van)
I've looked around extensively to see if I could find it in a paper, but
I don't see anything quite that on point. Van's classic paper,
"Congestion Avoidance and Control", does contain some figures (10 and 11)
which relate to it, and there's a lot of material which makes something
of the same point in "Promoting the Use of End-to-End Congestion Control
in the Internet" by Sally Floyd and Kevin Fall.
You need to note the qualifier "almost always" in the original statement.
Yes, there are some limited circumstances where the statement doesn't hold;
e.g. in which if a link is being used by one "aggressive" user, along with a
number of normal users, the aggressive user can grab more share. However, if
you have a number of such agressive users sharing a link, these strategies
E.g. several non-responsive flows cause each other to lose packets (the
Floyd/Fall paper has a good example of this); multiple TCP connections
cause the total link goodput to fall if everyone does it (if nothing
else, because of extra connection overhead); etc, etc.
More information about the end2end-interest