Re: Preserving established communications (was RE: about draft-nordmark-multi6-noid-00)

[Iljitsch van Beijnum <iljitsch@muada.com>]

On 27 okt 2003, at 15:45, marcelo bagnulo wrote:

> But i do have some simulation about how would the bgp detection 
> mechanism
> would affect an ongoing voip communication ;-)
> Considering a sample every 20 msec, that would be 3000 samples lost.

If we assume for a few moments that we can't make failover seamless, 
what would be an acceptable delay for VoIP?

Applications such as streaming media have the interesting property that 
the arrival of new data is extremely predictable. This makes it 
possible to use negative acknowledgements to trigger failover rather 
than the lack of regular acknowledgements, which is what TCP does.

If you expect a sample every 20 ms and you didn't get one for 100 ms, 
you can be pretty sure that something bad is happening, so it might be 
a good idea to signal the other side to try another address. This 
should give you failover within a fraction of a second without having 
to send any packets during regular communication. (Note that you still 
need ACKs in order to detect a correspondent going away altogether so 
you're not firing off packets into a black hole.)