Overlay routing meshes can be hard to debug, due to their decentralized nature, high concurrency, non-determinism, and large scale. We are developing tools, based on deterministic replay and post-mortem analysis, that improve the programmer's ability to examine real errors that occur in deployed networks. In this talk, I will present our work to date and discuss current problems.

The research described in this talk was done with Gautam Altekar, Ion Stoica and Scott Shenker.