Acceleration of Byzantine Fault Tolerance by Parallelizing Consensuses

摘要

We propose a new method that accelerates existing Byzantine Fault Tolerance (BFT) protocols for asynchronous distributed systems by parallelizing the involved consensuses. BFT realizes a reliable system against Byzantine failures and is usually solved by repeatedly executing consensus for a set of requests. Our method consistently parallelizes the consensus by introducing a new extra consensus on the order of processing agreed requests. We show the correctness of our method and analyze its performance in comparison with an existing non-parallelizing method and a naively parallelizing method. The results indicate that our parallelizing method is approximately 20% faster than those methods in such configurations where many replicas are running in order to increase reliability.