Leadership Uniformity in Raft Consensus Algorithm

The Raft consensus algorithm constitutes a widely-used algorithm not only in the broader area of distributed systems, ut also in private/permissioned blockchains such as Hyperledger Fabric. A Raft-based distributed system (RDS) strongly relies on leader election, which involves a number of time-related parameters. In the Raft-related literature, the process according to which those parameters are set is an under-researched area. Specifically, the use of the uniform distribution is the dominant approach. Motivated by this realization, in this work, we focus on these time parameters proposing the notion of “leadership uniformity” in combination with a series of performance metrics. Leadership uniformity is based on the desirable characteristic of having equality among the nodes who serve as leaders. The proposed performance metrics are straightforward adaptations of widely-used measurements from broad disciplines such as estimation theory. The experimental results of this work justify the appropriateness of the proposed notion of leadership uniformity. Specifically, the best performance was yielded by the utilization of normal distribution from which the time parameters under investigation were drawn.