RAFTing Over on Geo-Diverse Spot markets

摘要

Raft is a networked service used to synchronize cloud workloads at different locations. It is widely used, especially by workloads that span geographically distributed sites. As these workloads grow, Raft's costs should grow proportionally. However, auto scaling approaches for Raft inflate costs by provisioning at all sites when one site exhausts its local resources. This paper presents a Raft implementation, i.e., gRaft, that enables precise auto scaling on spot markets. gRaft extends Raft with the following abstractions: (1) secretaries which relieve log processing from leader nodes and (2) observers which relieve read requests from followers. These abstractions are stateless, allowing for elastic auto scaling, even with unreliable spot instances. gRaft preserves strong consistency guarantees provided by Raft. We set up and tested gRaft with multiple auto scaling techniques using traces from Google. gRaft scales in resource footprint increments 5-7X smaller than Multi-Raft, the state of the art. Using spot instances, gRaft reduces costs by 84.5% compared to Multi-Raft. gRaft improves goodput of 95th-percentile SLO by 9X.