Performance models of concurrency control protocols for transaction processing systems

Transaction processing plays a key role in a lot of IT infrastructures. It is widely used in a variety of contexts, spanning from database management systems to concurrent programming tools. Transaction processing systems leverage on concurrency control protocols, which allow them to concurrently process transactions preserving essential properties, as isolation and atomicity. Performance is a critical aspect of transaction processing systems, and it is unavoidably affected by the concurrency control. For this reason, methods and techniques to assess and predict the performance of concurrency control protocols are of interest for many IT players, including application designers, developers and system administrators. The analysis and the proper understanding of the impact on the system performance of these protocols require quantitative approaches. Analytical modeling is a practical approach for building cost-effective computer system performance models, enabling us to quantitatively describe the complex dynamics characterizing these systems. In this dissertation we present analytical performance models of concurrency control protocols. We deal with both traditional transaction processing systems, such as database management systems, and emerging ones, as transactional memories. The analysis focuses on widely used protocols, providing detailed performance models and validation studies. In addition, we propose new modeling approaches, which also broaden the scope of our study towards a more realistic, application-oriented, performance analysis.