An improved causal consistency mechanism for geo-replicated data stores

Reliable distributed data stores are an important pillar of today’s large-scale online systems. In order to accommodate the strict requirements imposed, careful trade-offs between consistency, availability and partition tolerance are made. Often data stores that employ a weaker consistency model, but have higher availability guarantees, are used to provide an ”alwayson” experience. However, this can expose the user to potential inconsistencies, the mitigation of which requires more complex application logic and tweaking. This paper presents a new causal consistency mechanism for key-value data stores that achieves performance and availability comparable to ones using eventual consistency. The idea is to preserve the virtues of strongly consistent storage services [1], while supporting incremental scalability and being highly decentralized and heterogeneous [2]. The work builds upon the design of Lloyd et al. [3], achieving near linear scalability with respect to the number of servers used per datacenter.