A Survey of Traditional and Practical Concurrency Control in Relational Database Management Systems

Abstract

Traditionally, database theory has focused on concepts such as atomicity and serializability, asserting that concurrent transaction management must enable correctness above all else. Textbooks and academic journals detail a vision of unbounded rationality, where reduced throughput because of concurrency protocols is not of tremendous concern. This thesis seeks to survey the traditional basis for concurrency in relational database management systems and contrast that with actual practice. SQL-92, the current standard for concurrency in relational database management systems has defined isolation, or allowable concurrency levels, and these are examined. Some ways in which DB2, a popular database, interprets these levels and finesses extra concurrency through performance enhancement are detailed. SQL-92 standardizes de facto relational database management systems features. Given this and a superabundance of articles in professional journals detailing steps for fine-tuning transaction concurrency, the expansion of performance tuning seems bright, even at the expense of serializabilty. Are the practical changes wrought by non-academic professionals killing traditional database concurrency ideals? Not really. Reasoned changes for performance gains advocate compromise, using complex concurrency controls when necessary for the job at hand and relaxing standards otherwise. The idea of relational database management systems is only twenty years old, and standards are still evolving. Is there still an interplay between tradition and practice? Of course. Current practice uses tradition pragmatically, not idealistically. Academic ideas help drive the systems available for use, and perhaps current practice now will help academic ideas define concurrency control concepts for relational database management systems.