[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [ns] Problem detected with TCP and bottleneck links



On Sun, 7 May 2000, Julian Waldby wrote:

> When the queue at the bottleneck link ($bottleQueueSize in attached
> code) is of size 16, TCP performs as expected. However, when the queue
> size is increased to 17, suddenly TCP performs in an erroneous fashion -
> the congestion window grows without bound.

The one-way tcp leaves slow-start at the window_ value (default and in
this case is 20) and doesn't attempt to simulate a buffer limit in the
sender or the receiver. If you're not dropping packets in the
queue of network, the window just keeps growing.

FullTcp behaves differently.

This isn't that much like real life; see e.g.:
ftp://ftp.ietf.org/internet-drafts/draft-handley-tcp-cwv-02.txt
(informational.)

L.

<[email protected]>PGP<http://www.ee.surrey.ac.uk/Personal/L.Wood/>

> Please try the following TCL script with
> "set bottleQueueSize 16"
> and
> "set bottleQueueSize 17"
> 
> I find that with the larger queue, TCP's congestion window grows
> steadily, never backing off.  Eg, when $totalTime is  3000.0 seconds,
> cwnd_ grows to more than 700.  Code is attached.  It uses standard
> building blocks of ns-2.1b6.
> 
> Julian
> --
> http://www.comm.csl.uiuc.edu/~waldby/