PhD defence by Yijian Liu
Title
Scalable and Transactional Actor-Oriented Databases
Abstract
In modern system design, the” separation of concerns” principle has led to two significant architectural trends: a shift from stateless to stateful middle tiers in the three-tier client-server architecture and the growing adoption of modular, loosely coupled architectures for complex applications. These trends enhance system responsiveness, scalability, and flexibility but also present challenges. The rising popularity of stateful middle-tier architectures, which maintain application states close to application logic in the middle-tier servers, offers numerous benefits but transfers the complexity of state management from backend databases to the application layer. In addition, it is particularly challenging to manage distributed states across loosely coupled components. The actor model [4] has experienced a resurgence as a tool for building stateful middle tiers and modular systems. By breaking down application logic into independent actors that communicate asynchronously, the actor model simplifies concurrency, scalability, and system resource management. However, its adoption has been limited due to the absence of critical database features such as transaction management, data replication, and data constraint enforcement. In response, the Actor-Oriented Database (AODB) [30] concept has emerged, proposing integrating these database capabilities into actor-based systems.
This dissertation emphasizes the need for further research and development in AODB. Specifically, we propose a scalable and transactional AODB to address the state management challenges for actor-based stateful middle tiers of modern applications. This dissertation achieves the goal through three steps: developing a transaction library for multi-actor transactions, designing a distributed system architecture, and introducing a data model to enable finer-grained state management. The outcome of this dissertation is a fully developed AODB featuring a clear and expressive programming model, a scalable actor-oriented architecture, efficient transaction processing techniques, a prototype implementation built on the Orleans framework, and comprehensive cloud-based evaluation.
Supervisors
Principal Supervisor Yongluan Zhou
Assessment Committee
Professor Philippe Bonnet, Department of Computer Science
Associate Professor Pinar Tözün, IT University of Copenhagen
Professor Xuan Zhou, East China Normal University
Leader of defense: Philippe Bonnet
It-responsible: Luiz Picoli
For an electronic copy of the thesis, please visit the PhD Programme page.