Byzantine-Resistant Total Ordering Algorithms

Abstract

Multicast group communication protocols are used extensively in fault-tolerant distributed systems. For many such protocols, the acknowledgments for individual messages define a causal order on messages. Maintaining the consistency of information, replicated on several processors to protect it against faults, is greatly simplified by a total order on messages. We present algorithms that incrementally convert a causal order on messages into a total order and that tolerate both crash and Byzantine process faults. Varying compromises between latency to message ordering and resilience to faults yield four distinct algorithms. All of these algorithms use a multistage voting strategy to achieve agreement on the total order and exploit the random structure of the causal order to ensure probabilistic termination.