Spectrum: A Framework for Adapting Consensus Protocols

Abstract

There exists a plethora of consensus protocols in literature. The reason is that there is no one-size-fits-all solution, since every protocol is unique and its performance is directly tied to the deployment settings and workload configurations. Some protocols are well suited for geographical scale environments, e.g., leaderless, while others provide high performance under workloads with high contention, e.g., single leader-based. Thus, existing protocols seldom adapt to changing workload conditions. To overcome this limitation, we propose Spectrum, a consensus framework that is able to switch consensus protocols at run-time, to enable a dynamic reaction to changes in the workload characteristics and deployment scenarios. With this framework, we provide transparent instantiation of various consensus protocols, and a completely asynchronous switching mechanism with zero downtime. We assess the effectiveness of Spectrum via an extensive experimental evaluation, which shows that Spectrum is able to limit the increase of the user perceived latency when switching among consensus protocols.