Design and Validation of the P-Store Replicated Data Store in Maude

Introduction. Many large applications—such as Google search, Gmail, Facebook, Dropbox, eBay, online banking and card payment processing—are expected to be available continuously, even under peak load, congestion in parts of the network, server failures, and during scheduled hardware or software upgrades. Such applications also typically manage large amounts of data. To achieve the desired availability, the data must be replicated across geographically distributed sites, and to achieve the desired scalability and elasticity, the data store may have to be partitioned across multiple partitions. It is well known [2] that it is hard or impossible to both guarantee strong correctness properties (such as serializability), high availability, and strong fault tolerance. The lack of strong correctness guarantees is acceptable for applications such as Google search, Facebook, and online newspapers, but is unacceptable in, e.g., online banking, online commerce (eBay and airplane reservation systems), and medical information systems. P-Store [6] is a well-known replicated and partitioned data store that provides both serializability and some fault tolerance (e.g., transactions can be validated even when some nodes participating in the validation are down). Wide-area replicated data stores designs are typically evaluated on real implementations or by using simulation tools, both of which are laborious tasks which cannot check “corner cases” to guarantee the absence of errors. In this talk I will talk about the use of the rewriting-logic-based Maude language and tool [3] to formally specify and analyze P-Store.