Bchain: Byzantine replication with high throughput and embedded reconfiguration

In this paper, we describe the design and implementation of BChain, a Byzantine fault-tolerant state machine replication protocol, which performs comparably to other modern protocols in fault-free cases, but in the face of failures can also quickly recover its steady state performance. Building on chain replication, BChain achieves high throughput and low latency under high client load. At the core of BChain is an efficient Byzantine failure detection mechanism called re-chaining, where faulty replicas are placed out of harm’s way at the end of the chain, until they can be replaced. Our experimental evaluation confirms our performance expectations for both fault-free and failure scenarios. We also use BChain to implement an NFS service, and show that its performance overhead, with and without failures, is low, both compared to unreplicated NFS and other BFT implementations.