RaftOptima: An Optimised Raft With Enhanced Fault Tolerance, and Increased Scalability With Low Latency

Abstract

In cloud computing and distributed databases, ensuring strong consistency and system reliability of the leader depends on distributed consensus algorithms such as Raft and Paxos. Raft is particularly appreciated for its simplicity and ease of understanding, making it a go-to choice for contemporary distributed systems. However, it faces challenges in efficiency and reliability, especially during network interruptions and when scaling up. The growing dependence on these consensus protocols for modern applications calls for improvements to address these issues. This paper presents RaftOptima, an advanced version of the Raft algorithm, which incorporates proxy leaders into the system to enhance scalability and introduces robustness against network failures. RaftOptima distributes the tasks of command distribution and gathering responses to proxy leaders, aiming to mitigate Raft's scalability problems and efficiency bottlenecks. Through simulations and TLA+ verification, RaftOptima showed 60% reduction in latency in configurations with up to 25 servers, highlighting its ability to significantly improve performance. Our improvements also increased the resilience of the Raft algorithm against partial network issues. This adaptation not only enhances Raft's operational effectiveness, but also its applicability and reliability in distributed systems for modern computing demands, meeting the critical demand for better scalability and liveness.