[e2e] Why do we need congestion control?

Matt Mathis mattmathis at google.com
Tue Mar 5 09:03:28 PST 2013

There is no disagreement as to the power and importance of statistical
multiplexing and it's need for some e2e congestion control.

It is just that the congestion control problem has be overly
constrained.  In particular I would place simpler requirements on the
end systems (no wasted capacity at high load, so total goodput equals
capacity) and then move some responsibility for the fairness/capacity
allocation problem to the network.

The "simple network, TCP friendly transport" paradigm does not work
for "fair" capacity allocation at scale and for any useful definition
of fair.

The best way to predict the future is to create it.  - Alan Kay

Privacy matters!  We know from recent events that people are using our
services to speak in defiance of unjust governments.   We treat
privacy and security as matters of life and death, because for some
users, they are.

On Tue, Mar 5, 2013 at 6:04 AM, Srinivasan Keshav <keshav at uwaterloo.ca> wrote:
> To answer this question, I put together some slides for a presentation at the IRTF ICCRG Workshop in 2007 [1]. In a nutshell, to save costs, we always size a shared resource (such as a link or a router) smaller than the sum of peak demands. This can result in transient or persistent overloads, reducing user-perceived performance. Transient overloads are easily relieved by a buffer, but persistent overload requires reductions of source loads, which is the role of congestion control. Lacking congestion control, or worse, with an inappropriate response to a performance problem (such as by increasing the load), shared network resources are always overloaded leading to delays, losses, and eventually collapse, where every packet that is sent is a retransmission and no source makes progress. A more detailed description can also be found in chapter 1 of my PhD thesis [2].
> Incidentally, the distributed optimization approach that Jon mentioned is described beautifully in [3].
> hope this helps,
> keshav
> [1] Congestion and Congestion Control, Presentation at IRTF ICCRG Workshop, PFLDnet, 2007, Los Angeles (California), USA, February 2007.
> http://blizzard.cs.uwaterloo.ca/keshav/home/Papers/data/07/congestion.pdf
> [2] S. Keshav, Congestion Control in Computer Networks PhD Thesis, published as UC Berkeley TR-654, September 1991
> http://blizzard.cs.uwaterloo.ca/keshav/home/Papers/data/91/thesis/ch1.pdf
> [3] Palomar, Daniel P., and Mung Chiang. "A tutorial on decomposition methods for network utility maximization." Selected Areas in Communications, IEEE Journal on 24.8 (2006): 1439-1451.
> http://www.princeton.edu/~chiangm/decomptutorial.pdf

More information about the end2end-interest mailing list