In supercomputers, there are such mechanisms for reactive congestion management: messages are sent to emitters to temporarily reduce flow. It works at small net

In supercomputers, there are such mechanisms for reactive congestion management: messages are sent to emitters to temporarily reduce flow. It works at small network scale but doesn't scale up, because congestion can happen often and for short times, and the latency of reaction is similar to the latency of events means you are not reacting to congestion anymore, you're just reducing global bandwidth. To this day the best practical mitigation strategies are great deterministic load balancing, QoS through virtual channels and fine-tuned adaptive routing. The analogy with traffic doesn't really hold though, because we can theoretically propagate throttling instructions faster than the vehicles involved, and the problem lies more with getting instructions applied at all. Just thought giving some overview of a similar problem would give some food for thought.