An Optimized Key-Value Raft Algorithm for Satisfying Linearizable Consistency

Xiao Liu, Zhao Huang, Quan Wang, Nan Luo

Abstract: Nowadays, the distributed systems have been widely applied in a variety of fields. However, this raises more concerns on reliability. Consensus algorithm is an important measure to ensure reliability of distributed systems, but its strong consistency may reduce the performance, resulting in cluster failure or even downtime. To this end, we propose an accelerated log backtracking optimization Raft algorithm, called ALB-Raft. It can improve the performance of traditional raft algorithm by enabling the backward tracker to update quickly. In particular, to achieve strong consistency, we construct a fault-tolerant distributed key-value (KV) service which conforms to the linearizable semantics. The experimental results illustrate that, when compared to the traditional raft algorithm, the proposed ALB-Raft consensus algorithm can resolve 20% of hundreds of log entry conflicts. Moreover, the ALB-Raft algorithm can also avoid the linear increase in the number of messages with the aggravation of log conflicts to ensure strong consistency.