Achieving High Availability of Consistent Data in the Presence of Network Partitions

摘要

We discuss availability maximization of a networked service that provides data access in a consistent and reliable manner. Majority voting is a common mechanism that implements such a service in the presence of network partitioning. A distinguishing feature of majority voting is that its availability is tunable by means of vote assignment to server nodes. We argue that achieving maximum availability of a voting-based service requires a powerful combinatorial optimization solving method and a new model of networks that can account for failure-repair behaviors observed in real networks. We suggest that the use of SAT/SMT-based optimization solving is a reasonable approach to combinatorial optimization of our interest. We also suggest the possible directions of augmentation of network reliability models.