An Optimized Raft Protocol Combined with Redundant Residue Number System

Abstract

Raft protocol requires 2F + 1 nodes to tolerate the F nodes failure. The leader needs to replicates complete log entries to F followers. As the number of nodes increases, the leader will generate a communication bottleneck, resulting in a decrease in log replication efficiency. In this paper, Raft protocol and Redundant Residue Number System are combined to change the way that Raft protocol replicates log entries. The leader replicates the log fragments encoded by the Redundant Residue Number System to followers. This method not only ensures the fault tolerance of Raft protocol, but also reduces the leader bandwidth consumption and the log entries storage overhead. The experiment shows that improved adRaft protocol based on Raft protocol reduces the leader bandwidth consumption by 23.39%, and the log entries storage overhead is reduced by 35.67%. Replicating the same number of log entries also takes 32.43% less time than the original Raft protocol. As more log fragments are generated by a log entry, the leader bandwidth consumption and log storage overhead decrease.