Using weaker consistency models with monitoring and recovery for improving performance of key-value stores
Journal of the Brazilian Computer Society
Consistency properties provided by most key-value stores can be classified into sequential consistency and eventual consistency. The former is easier to program with but suffers from lower performance whereas the latter suffers from potential anomalies while providing higher performance. We focus on the problem of what a designer should do if he/she has an algorithm that works correctly with sequential consistency but is faced with an underlying key-value store that provides a weaker (e.g.,
... tual or causal) consistency. We propose a detect-rollback based approach: The designer identifies a correctness predicate, say P, and continues to run the protocol, as our system monitors P. If P is violated (because the underlying key-value store provides a weaker consistency), the system rolls back and resumes the computation at a state where P holds. We evaluate this approach with graph-based applications running on the Voldemort key-value store. Our experiments with deployment on Amazon AWS EC2 instances show that using eventual consistency with monitoring can provide a 50-80% increase in throughput when compared with sequential consistency. We also observe that the overhead of the monitoring itself was low (typically less than 4%) and the latency of detecting violations was small. In particular, in a scenario designed to intentionally cause a large number of violations, more than 99.9% of violations were detected in less than 50 ms in regional networks (all clients and servers in the same Amazon AWS region) and in less than 3 s in global networks. We find that for some applications, frequent rollback can cause the program using eventual consistency to effectively stall. We propose alternate mechanisms for dealing with re-occurring rollbacks. Overall, for applications considered in this paper, we find that even with rollback, eventual consistency provides better performance than using sequential consistency. requirement of any system, network partition tolerance is a necessity. Hence, it is inevitable to make tradeoffs between availability and consistency, resulting in a spectrum of weaker consistency models such as causal consistency and eventual consistency [1,       . Weaker consistency models are attractive because they have the potential to provide higher throughput and higher customer satisfaction. On the other hand, weaker consistency models suffer from data conflicts. Although such data conflicts are infrequent , such incidences will affect the correctness of the computation and invalidate subsequent results.