HIGH PERFORMANCE DISTRIBUTED TRANSACTIONS FOR MULTI-REGION DATABASE SYSTEMS

Abstract

The growing popularity of global applications, such as social networks, online gaming, and supply chain management, has spurred the demand for a more advanced database system layer in multi-region cloud infrastructures. This layer should provide a strongly consistent interface, abstracting the complexity of distributed computing so developers can focus on application logic without worrying about data race issues. At the same time, it must leverage geographic redundancy to tolerate failures, ranging from single-node crashes to regional outages. Achieving these goals often conflicts with the need for scalability and high performance. Ensuring strong data consistency requires additional coordination between nodes, while enhancing availability and durability requires data replication. These measures increase latency and resource consumption, thereby limiting performance and scalability. As a result, existing solutions, such as NoSQL database systems, often compromise with weaker consistency levels, while fully geo-replicated systems incur performance and scalability penalties to achieve stronger consistency guarantees.

This dissertation addresses these challenges in three directions. First, we take a retrospective approach by enhancing an existing production-ready shared-storage architecture, derived from the Amazon Aurora database system. Our proposed modifications enables Aurora-style systems to support multiple writer nodes across geographically dispersed regions. Evaluation of our implementation, SunStorm, demonstrates substantial reductions in latency and improved scalability. Second, we adopt a forward-looking perspective by investigating a more recently proposed architecture known as deterministic database systems. We develop a new protocol within this architecture that facilitates the processing of strictly serializable multi-region transactions. Our implementation, Detock, achieves minimal performance degradation under high-contention workloads, improving throughput by an order of magnitude compared to state-of-the-art approaches and reducing latency by up to a factor of five. Finally, we conduct a comprehensive empirical study across a diverse range of real-world applications---such as e-commerce, chat, blogs, and content management systems---reassessing the assumptions of recent database system proposals and providing valuable insights for future transactional database research.