Byzantine Fault Tolerance Can Be Fast
- Miguel Castro ,
- Barbara Liskov
International Conference on Dependable Systems and Networks (DSN'01) |
Published by IEEE
Byzantine fault tolerance is important because it can be used to implement highly-available systems that tolerate arbitrary behaviorfrom faulty components. This paper presents a detailed performance evaluation of BFT, a state-machine replication algorithm that tolerates Byzantine faults in asynchronous systems. Our results contradict the common belief that Byzantine fault tolerance is too slow to be used in practice – BFT performs well so that it can be used to iniplenient real systems. We implemented a replicated NFS file system using BFT that performs 2% faster to 24% slower than production implementations of the NFS protocol that are not fault-tolerant.