A comparison of fault-tolerant atomic broadcast protocols

Abstract

User expectations for dependable distributed computing follow a growing trend. Atomic broadcast is at the core of problems such as reliable commit, atomic transactions, and group membership that must be tackled in distributed systems in the presence of various types of failures. However, for many applications, the usual specification of fault-tolerant atomic broadcast is inadequate, as it does not define the behaviour of faulty processors. The purpose of this work is to present the reason why previously published algorithms fail to achieve the properties sought (e.g. atomicity, total order). Hence, eight of the most significant atomic broadcast protocols are analyzed according to five properties which appear to be necessary in a reliable distributed system.<>