[Iccrg] What's wrong with TCP Congestion control

[Phelan, Tom]

Hi All,

I'd like to point out that the large buffer requirement is not an
intrinsic property of window-based congestion control.  It's the result
of the requirement for reliability, coupled with the large transmit
window.  If you don't need reliability (as with DCCP and its AIMD
congestion control), the need for sender buffering is greatly reduced.

Tom Phelan

> -----Original Message-----
> From: iccrg-bounces at cs.ucl.ac.uk [mailto:iccrg-bounces at cs.ucl.ac.uk]
On
> Behalf Of Simon Leinen
> Sent: Tuesday, February 07, 2006 8:43 AM
> To: Radek Krzywania
> Cc: iccrg at cs.ucl.ac.uk
> Subject: Re: [Iccrg] What's wrong with TCP Congestion control
> 
> Radek Krzywania writes:
> > On Tue, 07 Feb 2006 01:50:29 +0100, S. Keshav <keshav at uwaterloo.ca>
> wrote:
> >> Here is a list paraphrased from mail so far:
> >>
> >> 1. Unsustainable large equilibrium window sizes under high
> >> bandwidth-delay
> >> product environments; requires an unrealistically low loss
probability.
> >> (Leads to limited dynamic range)
> 
> > Maybe it is good to say explicitly, that large windows requires
large
> > memory buffers on sender and receiver sides.
> 
> The large buffer requirements are mostly due to the fundamental nature
> of window-based transmission, independent of TCP's congestion control
> algorithm: You need to support a large amount of unacknowledged data
> (the window) to sustain the throughput, and you need the buffers at
> the sender so that you can retransmit.
> 
> (The requirement for equally large buffers at the receiver seems more
> like a constraint of the current prevailing interface to TCP in
> end-systems.)
> 
> The only relation between TCP's congestion control and required buffer
> sizes that I can see is when queueing at the bottleneck starts to
> significantly increase RTT.
> --
> Simon.
> 
> 
> _______________________________________________
> Iccrg mailing list
> Iccrg at cs.ucl.ac.uk
> http://oakham.cs.ucl.ac.uk/mailman/listinfo/iccrg