PERFORMANCE MEASUREMENT, MODELLING, AND EVALUATION OF INTEGRATED CONCURRENCY CONTROL AND RECOVERY ALGORITHMS IN DISTRIBUTED DATABASE SYSTEM

In this dissertation, the performance evaluation of integrated concurrency control and recovery mechanisms for distributed database systems is studied using a distributed testbed system. In addition, a queueing network model has been developed to analyze the two-phase locking scheme in the distributed testbed system. The combination of testbed measurement and analytical modelling provides an effective tool for understanding the performance of integrated concurrency control and recovery algorithms in distributed database systems.;The design and implementation of the distributed testbed system, CARAT, are presented. The concurrency control and recovery algorithms implemented in CARAT include: a two-phase locking scheme with distributed deadlock detection, a distributed version of optimistic approach, before-image and after-image journalling mechanisms for transaction recovery, and a two-phase commit protocol.;Many performance measurements have been conducted using a variety of workloads. In the workloads, there are four types of transactions: local read-only, local write, distributed read-only, and distributed write. The performance measures of the two-phase locking scheme with before-image journlling and the optimistic approach with after-image journalling are presented and analyzed. The major performance measures include transaction throughput, probability that a transaction is aborted, CPU utilization, and disk I/O rate. Basic parameter values, such as locking overhead, mean lock blocking time, and validation overhead, are also measured.;A queueing network model is developed to analyze the performance of the CARAT system using the two-phase locking scheme with before-image journalling. The queueing model differs from previous analytical models in three major aspects. First, it is a model for distributed transaction processing. Second, it is more general and integrated than previous analytical models. Finally, this model reflects a functioning distributed database testbed system and is validated against performance measurements.;The combination of testbed measurements and analytical modelling provides significant improvements in understanding the performance impacts of the concurrency control and recovery algorithms in distributed database systems.