Research on Consensus Efficiency Based on Practical Byzantine Fault Tolerance

Abstract

Byzantine Fault Tolerance (BFT) will be increasingly important to provide reliability in distributed systems and various Byzantine fault-tolerant algorithms have been proposed to improve efficiency. However, the practicality of BFT is still limited by its cost of consensus. Practical Byzantine Fault Tolerance (PBFT) is one of the best algorithms used in reality. Confronted with a distributed system with numerous nodes, PBFT is not satisfying. In order to improve the consensus efficiency, a new algorithm named Group-Hierarchy (GH) based on PBFT is proposed in this article. GH divides all replicas in a distributed system into several groups and every group has a primary. Replicas in a group reach an agreement respectively, which is called Local Consensus. And primaries of all the groups also reach a consensus based on the Local Consensus, which is called Global Consensus. After the Local Consensus and Global Consensus, replicas could reply to client. GH correlates the message complexity to the number of replicas and the number of groups in a distributed system in theory. The message complexity of GH is lower than PBFT in theory and the performed tests show that GH can execute a request even several times faster than PBFT.