Re: TCP (and SCTP) sucks on high speed networks

[Mark Allman <mallman@grc.nasa.gov>]

> On Fri, 1 Dec 2000, Dawkins, Spencer wrote:
> 
> > RFC 2581 calls for an increase in the congestion window for each
> > RTT's worth of acks, not for each ack, during congestion
> > avoidance.
> 
> The exact opposite, surely?
> 
> Below, 'This adjustment is executed on every incoming non-duplicate
> ACK. ' is a Dead Giveaway. The adjustments per ack approximate to an
> adjustment per RTT...
> 
> unless you're not counting fractions, and 'an increase' in your
> message == 'add one'.

Just to really beat this to death, I think Spencer is right.  The
idea in the RFC is, indeed, "one segment per RTT".  The sample
implementation he cited from the RFC is, indeed, executed per ACK.
However, to follow the spirit of the RFC that doesn't necessarily
need to be the case.  For instance, the RFC provides another option
for increasing cwnd during congestion avoidance...

   Another acceptable way to increase cwnd during congestion avoidance
   is to count the number of bytes that have been acknowledged by ACKs
   for new data.  (A drawback of this implementation is that it requires
   maintaining an additional state variable.)  When the number of bytes
   acknowledged reaches cwnd, then cwnd can be incremented by up to SMSS
   bytes.

allman


---
Mark Allman -- NASA GRC/BBN -- http://roland.grc.nasa.gov/~mallman/