Mixed Real-Time Transaction Scheduling And Concurrency Control

In real-time database systems, transactions with different types of deadlines have different scheduling requirements. For example, a hard real-time transaction, which misses its deadline, may result in great loss even catastrophe. But it is tolerable for a small quantity of soft real-time transactions to miss their deadlines. Previous study mainly focuses on scheduling and concurrency control on either hard or soft real-time transactions. However, there is an increasing demand in processing mixed transactions in real-time database systems. The system goal with mixed real-time transactions is that the deadlines of hard real-time transactions must be guaranteed, and the number of deadline violations for soft real-time transactions should be minimized.In this paper, two new concurrency control protocols -Hard Concurrency Control with Dynamic Adjustment of serialization order (HCC-DA) and Hard Concurrency Control with Extended Dynamic Adjustability (HCC-EDA), are proposed for hard real-time transactions, which are scheduled under dynamic priority scheduling algorithms. Both theoretic analysis and simulation test show that, these two protocols can improve the schedulability of hard real-time transactions, and achieve higher processor utilization.Next, a framework for mixed transaction scheduling is given, and a new concurrency control protocol - Mixed Concurrency Control with Dynamic Adjustment of serialization order using Timestamp Intervals (MCC-DATI) under the above framework is proposed. The simulation results show that, the proposed protocol can guarantee the deadlines of hard real-time transactions, and reduce the number of deadline violations of soft real-time transactions.Furthermore, a feedback control mixed scheduling algorithm (FC-MTS) is proposed by applying feedback control theory to mixed transaction scheduling framework. The simulationresults show that the proposed algorithm can provide some performance guarantee for softreal-time transaction without affecting the schedulability of hard real-time transactions.In addition, a test platform, AgilorTP, is developed for evaluating the performance of proposed algorithms or protocols in real-time databases, which can provide full extendibility and configurability to support the test and analysis of new real-time transaction models, scheduling algorithms, concurrency control protocols, real-time I/O scheduling and buffer management policies.The work in this paper provides the foundation for further research on mixed real-time transaction scheduling and concurrency control, and processing mixed transactions in actual application systems.