An Improved Raft Protocol Combined with Cauchy Reed-Solomon Codes

Raft can provide highly reliable and available distributed services, it needs at least 2F+1 nodes to tolerate any F nodes failures. In Raft, the methods of leader replicates complete log entry to all followers causes high communication cost and storage cost. This harm the performance of Raft. In this paper, leader initiates a different log replication request to different followers, then each follower uses Cauchy Reed-Solomon codes to generate the same number of log data blocks as followers and sends them to other followers, this reduces the communication cost of leader and storage costs of clusters. Experiments show that the communication cost of leader is reduced by 73.86%, storage cost of cluster is reduced by 41.99% and request processing delay is reduced by 40.39%.