On 05/24/2012 01:02 PM, Zaheduzzaman Sarker wrote:
Hi,
One topic that so far have not been discussed in this mailing list is the ConEx and it's effect on congestion avoidance.
ConEx Background: IETF ConEx WG is chartered here https://datatracker.ietf.org/wg/conex/charter/. According to that charter ConEX is working on congestion exposure mechanism for IPv6 networks. Basically, the receiver of a flow sends the congestion related information (packetloss, ECN-CE markings) back to the sender then the sender of the flow inserts IPv6 header extension to expose that information to the operators. This is done to aid the congestion management at the network. Typically as long as the flow does not exceed a certain congestion volume the network does not do any kind of traffic shaping on that flow. Thanks for the heads-up!
I see that https://datatracker.ietf.org/doc/draft-ietf-conex-concepts-uses/ shows that one document is already in front of the IESG; is this a good starting point to read up on it?
The effects:
+ RTP media is not ConEx enabled: Bitrate may be throttled in the network possibly based on some time of day policy or whatever.
+ RTP media is ConEx enabled but no feedback or congestion information to the sender or congestion feedback is too slow : Audit functions will find a mismatch between stated and actual congestion and will start to drop packets.
What timescales are we talking about here - what's the time over which ConEx is going to expect congestion information to be accurate?
+ RTP media is ConEx enabled and timely feedback of congestion info to sender: Packets will pass through unaffected by the ConEx policers as long as congestion volume quota is not exceeded.
This means:
* the operators can drop priority on non-ConEx flows hence a ConEx enabled flow is treated differently. This will have impact on the congestion avoidance techniques RMCAT will produce as same algorithm may not work efficiently enough for both ConEx enabled flow and non-ConEx enabled flow.
* a ConEx enabled flow will need to send congestion related information (perhaps more frequently than usual) i.e. packet loss and ECN marking information along with simple rate request.
* RTP media need to be congestion volume aware.
I see a clear impact on design choice on how to handle these. I think we should discuss the impact of ConEx here before the BOF in Vancouver.
The list's been relatively quiet for a while, so there's room ... the result of discussion may be "too far out to worry about"; my 0.01 read is that Conex expects that it has to be deployed in both ISPs and endpoints before it does any good, while the goal for RMCAT is a function that can be usefully deployed even if it's deployed at the endpoints only. Harald