Review of RDMA-enabled consensus protocols

Abstract

Several cloud computing applications use Replicated State Machines (RSM) to provide fault-tolerant services, ensuring consistency with consensus protocols. However, these protocols often come with a high latency cost, sometimes even forcing system designers to sacrifice consistency for availability. This latency is due, in part, to unnecessary data copies in the kernel TCP/IP layers. Remote Direct Memory Access (RDMA) bypasses the kernel to provide faster communication and lower CPU overhead through zero-copy data transfer. Recent works have utilized RDMA primitives to improve the performance of consensus protocols. However, integrating RDMA into such protocols and utilizing it efficiently can be a complex task. In this paper, we address this problem by presenting a systematic review of the state-of-the-art approaches for implementing RDMA-based consensus protocols.