Brief Announcement

Abstract

Replication protocols in distributed storage systems are fundamentally constrained by the finite propagation speed of information, which necessitates trade-offs among performance metrics even in the absence of failures. We make two contributions toward a clearer understanding of such trade-offs. First, we introduce a probabilistic model of eventual consistency that captures precisely the relationship between the workload, the network latency, and the consistency observed by clients. Second, we propose a technique for adaptive tuning of the consistency-latency trade-off that is based partly on measurement and partly on mathematical modeling. Experiments demonstrate that our probabilistic model predicts the behavior of a practical storage system accurately for low levels of throughput, and that our tuning framework provides superior convergence compared to a state-of-the-art solution.