An Implementation of Attributive Predicate Lock in Database System

In todays application scenarios of OLTP database, transactions are generally made up of simple queries, especially ones based on primary keys. In such cases, the predicate lock based on semantics of simple comparisons can be used to solve the problem of Non-repeatable read and phantom read, which guarantees the serializable schedule of transactions, leaving out complex logical judgments. In consideration of the advantages of predicate lock, we propose an extended version called Attributive Predicate Lock, which supports modifications on different attribute domains in the same data row, thus improve the concurrent capabilities of transactions. We discuss the feasibility of this theory in the given complexity and OLAP application scenarios, and give a detailed implementation of such a lock system. Based on the theory and implementation, we also conduct a simulation of TPC-C concurrent transactions on the framework of transaction threads in the national ShenTong Database System, and compared its performance with that of physical lock system. The experiment results prove that Attributive Predicate Lock can schedule transactions correctly with better performance of CPU and memory in the scenario of simple queries and updates on fixed attribute columns, which will be of great improvement to the concurrent capabilities in OLAP applications.