RL-Paxos: Relieving the Leader's Burden with Efficient Task Offloading in Distributed Consensus

Chenhao Zhang

Abstract:

Existing state machine replication protocols are typically leader-based, with the leader responsible for all operations, leading to a scalability bottleneck. Despite extensive research efforts to address this issue, alleviating the leader's workload remains a significant challenge.
In this paper, we provide an in-depth analysis of the leader bottleneck and introduce RL-Paxos, which retains only the core function of ordering (e.g., establishing a global order and performing commutativity checks) for the leader, while offloading other tasks to the followers.
We integrate RL-Paxos into the classic Paxos protocol and compare it with state-of-the-art protocols. Extensive experimental evaluations demonstrate that RL-Paxos achieves high throughput and low wide-area latency under conditions of high concurrency, multi-node deployment, mixed workloads, and read-only scenarios.