Practical Byzantine Fault Tolerance
Barbara Liskov
Ford Professor of Engineering
Massachusetts Institute of Technology

The increasing reliance of industry and government on online information services makes malicious attacks on these systems more and more attractive and the consequences of such attacks are very serious. This talk will describe a new replication technique that allows services to withstand such Byzantine failures; the system is not only resilient to malicious attacks, but it also can continue to operate correctly in the presence of software bugs.

The new algorithm is of interest for a number of reasons. It is the first approach that allows correct functioning over the lifetime of the system provided the number of Byzantine faults occuring in some small time window (e.g., 5 minutes) is bounded. It supports general applications and works in an asynchronous environment such as the Internet. And it includes a number of important optimizations that allow it to perform well in practice.

The talk will describe the algorithm. It will also discuss the experimental results. The results show that the performance of a replicated version of NFS is competitive with the production implementations of NFS, which are used daily by many users and are not replicated.