[e2e] packet-pair probe implementation

[Rik Wade]

On Tue, 13 May 2003, Michael Welzl wrote:
> 
> So that's the bottleneck bandwidth ... the bottleneck capacity ...
> to nominal bottleneck link bandwidth ... whatever - but not what I
> meant when I said "available bandwidth". These terms are just
> confused too often.

Yes, sorry, my mistake. The email was written after a 3 hour drive at
4am, that's my excuse anyway. I have taken on board the distinction made
on-list over the past week and agree wholeheartedly with the conclusions. It
is indeed _capacity_ we are measuring here, not _bandwidth_.

> I also tend to believe that it's a good idea for a congestion control
> mechanisms to use packet pair. I like these approaches (LDA+, PLM, ..).

So like Keshav's original proposal for the Packet Pair protocol, do others
agree that taking regular samples of bottleneck capacity using a packet pair
mechanism is a good solution? i.e. incorporating the packet pair mechanism
in to the congestion avoidance framework? The work I did only used packet
pair for the protocol's initial startup in order to estimate bottleneck
capacity. A proactive congestion avoidance algorithm similar to that of 
TCP Vegas was then used to react to path congestion.

My track was to use packet pair for a "finger in the air" indication of 
path congestion during startup, but then to use something a bit more reliable 
(and more easily modelled ;-) for the ongoing congestion avoidance and 
steady-state operation of the transport protcol.

--
rik