An effective scheduling algorithm for parallel transaction processing systems

We study the scheduling problem in shared-nothing parallel transaction processing (TP) systems with two-phase locking concurrency control. A transaction scheduling algorithm called SCST (Synchronizing Completion of Sub-Transactions) is proposed to reduce the data contention and thus to improve the system performance in the shared-nothing parallel TP systems. Furthermore, a comprehensive simulator for shared-nothing parallel TP systems is developed for studying the performance behavior of the SCST algorithm. Using this simulator, we compare the SCST algorithm with the conventional FCFS (First-Come-First-Served) scheduling algorithm. We show that the SCST algorithm outperforms the FCFS algorithm significantly in terms of the transaction throughput, the 95th percentile of the response time, and the restart ratio.