Re: Q on the congestion awareness of routing protocols

[Stewart Bryant <[e-mail]>]

> On 2 Dec 2022, at 18:21, Matt Mathis <mattmathis=[e-mail]> wrote:
> 
> 
> That work was more than 30 years ago, so it might pay to revisit it, to re-examine it to see if something has changed.   However, the conclusion matches my intuition.

A few thoughts spring to mind, firstly routing normally works on the basis of truth. This means that everyone jumps to the shortest low delay path. You could ask routers to be altruistic and use SR to use a higher latency path for some of their flows. A more radical approach is to get some routers to lie about cost in some way to reduce the extreme.

Secondly it occurs that since we last looked at this there has been the development of shortest time routing in car Sat navigation and there might be a trick or two to be borrowed from that industry. For example cars have de facto multiple independent competing algorithms working in parallel. So we could again use SR with some form of local AI based path choice mechanism.

As Toerless says we really need a routing research area in the IETF to develop these sorts of ideas. We have tried and always been met with refusal from IRTF management.

One thought is that we set up a shadow RRG operating independently of the IETF meeting but meeting concurrently and at the same location. This is what BoFs used to be before the IETF apparatus got hold of them and formalised them. We can still use the draft infrastructure. If we cannot have a IETF list, there is groups.io or similar. All a bit like Internet Routing where we route around damage to the infrastructure.

Stewart