Fast and Reliable Byzantine Fault Tolerance

Abstract

Byzantine faults, or arbitrary faults, are difficult to handle due to their unknown nature. They include software errors, hardware errors, and malicious behavior. There are several algorithms which handle Byzantine faults with varying degrees of reliability and speed. The purpose of this thesis is to combine different Byzantine fault tolerant algorithms in a publisher/subscriber service in such a way that achieves high speed and reliability.

Faster algorithms are used for the majority of the publications, but after every α publications, a history publication is sent with a more reliable algorithm. The history publication will allow subscribers to learn any missed publications. Three algorithms are used: Authenticated Broadcast, Bracha’s Reliable Broadcast, and Chain. Bracha’s Reliable Broadcast is the most reliable of the three, and it is used in broadcasting the history publications.

By combining these algorithms through the use of the history publication, this thesis demonstrates that it is possible to have a distributed, Byzantine fault tolerant service that is both fast and reliable.