Concurrency Control Algorithms for Real-Time Database Systems
Lee, Juhnyoung, Department of Computer Science, University of Virginia
Son, Sang, Department of Computer Science, University of Virginia
In addition to satisfying data consistency requirements as in conventional database systems, concurrency control in real-time database systems must also satisfy timing constraints, such as deadlines associated with transactions. Concurrency control for a real-time database system can be studied from several different perspectives. This largely depends on how the system is specified in terms of data consistency requirements and timing constraints.
The objective of this research is to investigate and propose concurrency control algorithms for real-time database systems, that not only satisfy consistency requirements but also meet transaction timing constraints as much as possible, minimizing the percentage and average lateness of deadline-missing transactions.
To fulfill the goals of this study, we conduct our research in three phases. First, we develop a model for a real-time database system and study the performance of various concurrency control protocol classes under a variety of operating conditions. Through this study, we understand the characteristics of each protocol and their impact on the performance, and ensure the validity of our real-time database system model by reconfirming the results from previous performance studies on concurrency control for real-time database systems. Second, we choose optimistic technique as the basic mechanism for our study on concurrency control for real-time database systems, and investigate its behavior in a firmdeadline environment where tardy transactions are discarded. We present a new optimistic concurrency control algorithm that outperforms previous ones over a wide range of operating conditions, and thus provides a promising candidate for the basic concurrency control mechanism for real-time database systems. Finally, we address the problem of incorporating deadline information into optimistic protocols to improve their real-time performance. We present a new priority-cognizant conflict resolution scheme that is shown to provide considerable performance improvement over priority-insensitive algorithms, and to outperform the previously proposed priority-based conflict resolution schemes over a wide operating range. In each step of our research, we report the performance evaluation results by using a detailed simulation model of real-time database system developed in the first phase.
In addition to the three phases, we investigate semantic-based concurrency control techniques for real-time database systems, in which the semantics of operations on data objects are used to increase the concurrency of transactions executing on the data objects and to meet the timing constraints imposed on the transactions. We propose an object-oriented data model for real-time database systems. We present a semantic-based concurrency control mechanism which can be implemented through the use of the concurrency control protocols for real-time database systems studied earlier along with a general-purpose method for determining compatibilities of operations.
PHD (Doctor of Philosophy)
All rights reserved (no additional license for public reuse)