[AVT] RTP end-to-end liveness test

[Henning Schulzrinne <hgs@cs.columbia.edu>]

(This discussion was part of a thread on 2833, which most people 
probably ignored, but I believe it's worth discussing separately, 
without getting tied up with V.21 tones.)

Flemming had suggested a liveness test for RTP, where one side would 
send a test packet and get back an answer immediately if that test 
packet arrived. This would seem particularly useful where various NATs 
and firewalls make IP-level reachability tests inconclusive or fail. My 
concern is that the sender of the "echo request" has no sure way of 
knowing how many will respond. Implosion would not be limited to 
multicast, but could happen for a mixer-based configuration as well. To 
mitigate such concerns, one would effectively have to replicate the 
mechanisms in RTCP, i.e., track the session population, do 
reconsideration, etc.

I think this feature is worth discussing. I would find something like 
"send a test packet and then wait if the RTCP counters reflects that 
packet" to be almost as good, but without incurring the implosion 
problem. This probably doesn't even require anything new, since this 
would work by sending an empty RTP packet, say. Use of RTCP does impose 
a delay of up to a few seconds. An approach such as this has the 
advantage that it works for existing compliant implementations (i.e., 
ones that implement RTCP).

It will be hard to eliminate the RTCP delay without risking multicast 
implosion problems.

Henning

_______________________________________________
Audio/Video Transport Working Group
avt@ietf.org
https://www1.ietf.org/mailman/listinfo/avt