Scalable and Fully Consistent Transactions in the Cloud through Hierarchical Validation

Abstract

Cloud-based systems are expected to provide both high availability and low latency regardless of location. For data management, this requires replication. However, transaction management on replicated data poses a number of challenges. One of the most important is isolation: Coordinating simultaneous transactions in a local system is relatively straightforward, but for databases distributed across multiple geographical sites, this requires costly message exchange. Due to the resulting performance impact, available solutions for scalable data management in the cloud work either by reducing consistency standards (e.g., to eventual consistency), or by partitioning the data set and providing consistent execution only within each partition. In both cases, application development is more costly and error-prone, and for critical applications where consistency is crucial, e.g., stock trading, it may seriously limit the possibility to adopt a cloud infrastructure. In this paper, we propose a new method for coordinating transactions on replicated data. We target cloud systems with distribution across a wide-area network. Our approach is based on partitioning data to allow efficient local coordination while providing full consistency through a hierarchical validation procedure across partitions. We also present results from an experimental evaluation using Real-Time Maude simulations.

This work was partially supported by AFOSR Grant FA8750-11-2-0084.