Performance engineering of group communication protocols

Many complex applications can benefit from distributed systems based on multiple computers interconnected by a communication network. Distributed systems use inexpensive high-performance computers and can be configured to closely match the needs of the application. Information can be replicated on several processors to improve performance and to provide fault tolerance.; However, programming distributed applications is difficult, particularly when replicated information is being updated and must remain consistent in the presence of faults. Providing a total order on packets ensures that the same updates will be performed in the same order at the different application processes, allowing the application programmer to concentrate on the application rather than on the underlying communication protocols.; In the past, total-ordering protocols have exhibited poor performance. Proper engineering of such protocols can achieve high performance. This dissertation discusses several different approaches for achieving high performance in group communication systems, along with the advantages and drawbacks of each approach. The work is based on the Totem system, which uses careful protocol design and system engineering to provide high performance. Fine tuning of the protocol is required to achieve the best system performance. Performance graphs, in terms of packet throughput and latency to message delivery, for different network topologies and under different operating conditions, are presented. The different factors that contribute to Totem's high performance are also discussed in this dissertation.