Although Ensemble is a big project, some themes unify our work. One recent theme is concerned with probabilistic protocols and the general question of using probabilistic tools in distributed settings. Our interest in this topic stems from two observations. (1) Probabilistic protocols scale extremely well and have very steady data throughput under loads and conditions that cause other reliability models to falter. (2) Traditional reliability protocols may hang when systems experience soft, or undetected, failures; probabilistic ones continue to make progress.
A second theme is concerned with using formal methods to improve the quality. This work is being done jointly with R. Constable's Nuprl project and N. Lynch's group at MIT. The idea is to develop specifications for Ensemble, using I/O Automata, that can be used to prove the correctness of key aspects of our protocol stacks. We are doing this in a way that uses the Nuprl system as a high-powered tool for automating the necessary proofs and for transforming (optimizing) Ensemble to better match the environments in which the system is actually used and the patterns of communication seen in real-world situations.
The Ensemble project is directed by myself, R. van Renesse, and W. Vogels. B. Ban is visiting as a post-doc during 1998-2000. T. Clark manages our software distribution effort and generally keeps things running. During 1997-1998, PhD graduates included M. Hayden, A. Vaysburd, and K. Guo.
Editor-in-Chief: ACM Transactions on Computing Systems
The Next Generation Internet: Unsafe at Any Speed. Brown University, Feb. 1998.
Building Secure and Reliable Network Applications. Manning Publishing Company (Greenwich, CT) and Prentice Hall (1997). For information, http://www.browsebooks.com/Birman/index.html.