[e2e] High Packet Loss and TCP

[David G. Andersen]

On Thu, May 01, 2003 at 02:36:53PM -0700, Greg Minshall quacked:
> David,
> 
> i'm surprised by the 30% number as it seems high.  i guess my rule of thumb 
> has been that TCP might behave itself under 10%, but that over 10% you're 
> unlikely to get anything decent in terms of performance.
> 
> maybe this is the difference between theory and practice (which difference is 
> that in theory there is none, but in practice there is), or maybe the current 
> turbo-charged TCPs (fast retransmit, etc.) really bump the performance in a 
> lossy environment.

  You're right - I don't find TCP particularly useful at anything over
10% either.  The curve I cite in the paper shows an even worse reduction after
30%, but the performance under appreciable loss is terrible all along.

  They sent bulk data between various pairs of hosts for an hour.
At 30%, they were down to about 10-100 packets successfully received
in an hour.  At 1%, they got a couple hundred through;  by 40-50%, they
got about fewer than ten packets through.

  100 packets/hour equates to a (very bursty) ~330 bits/sec;  putting on
my wayback hat, using a 300 baud modem was barely tolerable for interacting
with text-based systems.  1200 baud (~10% loss) wasn't too bad, though
it'd be much worse when it happened in bursts separated by a few RTOs.

  -Dave

-- 
work: dga at lcs.mit.edu                          me:  dga at pobox.com
      MIT Laboratory for Computer Science           http://www.angio.net/
      I do not accept unsolicited commercial email.  Do not spam me.