MACEDON is a system for creating, evaluating, and designing overlay networks and distributed systems. This page describes Chip's personal research as part of the MACEDON team. For more information on the MACEDON project, visit it's homepage, which can be found here. (http://www.cs.duke.edu/~razor/MACEDON/)
I am currently working in two main areas. First in the area of measuring and evaluating existing overlay networks, and second in the area of parameter optimization in distributed systems.
Measuring and evaluating existing overlay networks: MACEDON provides an ideal framework for comparing overlay networks. Combined with ModelNet, these experiments share the vast majority of codebase, as well as being run over a predictable and moderately real network emulation (its main deficiencies are in  when the hardware is limited, and  lack of realistic internet cross-traffic). By first validating our implementations of algorithms, and then running tests over the same network on code with the same optimizations, these results are highly comparable.
Parameter optimization: Most protocols include several parameters which the authors explain need to be tuned given your circumstances. But seldom are guides given which explain how best the parameters should be tuned. It is even suspected that frequently authors do not know. But with MACEDON, this need not be so. I am currently developing a toolkit of techniques which can be applied generically to overlay algorithms. First, I am working on a set of machine or reinforcement learning techniques. As a proof of concept, I have coded a modified version of Chord (I call it Q-Chord), which uses Q-learning to determine the best time to perform its periodic routing updates. Second, I plan to implement optimization methods such as the empirical gradient, which can be used to automatically run experiments and to find the best values of the parameters given the running environment.