An Experimental Study of the Learnability of Congestion Control

When designing a distributed network protocol, typically it is in-feasible to fully define the target network where the protocol is in-tended to be used. It is therefore natural to ask: How faithfully do protocol designers really need to understand the networks they design for? What are the important signals that endpoints should listen to? How can researchers gain confidence that systems that work well on well-characterized test networks during development will also perform adequately on real networks that are inevitably more complex, or future networks yet to be developed? Is there a tradeoff between the performance of a protocol and the breadth of its intended operating range of networks? What is the cost of play-ing fairly with cross-traffic that is governed by another protocol? We examine these questions quantitatively in the context of con-gestion control, by using an automated protocol-design tool to ap-proximate the best possible congestion-control scheme given im-perfect prior knowledge about the network. We found only weak evidence of a tradeoff between operating range in link speeds and performance, even when the operating range was extended to cover a thousand-fold range of link speeds. We found that it may be ac-ceptable to simplify some characteristics of the network—such as its topology—when modeling for design purposes. Some other fea-tures, such as the degree of multiplexing and the aggressiveness of contending endpoints, are important to capture in a model