On flows of this size, I can see CAR producing better results. However
since it does no smoothing of flows, it produces nasty TCP performance,
as when customers burst, they will find their TCP sessions exceed their
excess burst and drop packets, causing TCP to go into slow restart.
In our applications (2meg LL shaped to 1meg and such) this is
not so useful. We have also evaluated, and are using the Packeteer,
which is I guess what you meant?
Cheers,
Chris.
On Mon, Dec 04, 2000 at 09:24:53AM +0000, Neil J. McRae wrote:
> > Generic traffic shaping would be the best thing I think. CAR isn't intended
> > for customer traffic shaping, and probably wouldn't work too well.
>
> Hmm, We use CAR for customer traffic management and it works very
> well, infact of all the stuff we tested CAR worked the best. We
> also looked at Packetshaper and Xedia. For high bandwidth, well
> oversubscribed customers [e.g. they buy 2M when they really need 10M],
> with large numbers of flows (such as free ISPs), CAR was the only
> thing that worked. [on 11.1CC branch].
>
> Neil.
-- |=========----- -------=======| Unix, BASIC, C, PASCAL, APL, ADA, | Chris Roberts (chrisr@pavilion.net)| and PROFANITY spoken here. |=======------- -----=========| T: 0845 333 5000 F: 0845 333 5001
This archive was generated by hypermail 2b29 : Sun Aug 04 2002 - 04:12:22 EDT