Research Of Main Memory Database Parallel Query Processing On SMP

Main Memory Database(MMDB) has a lot of advantage comparing to traditional database, and it uses a different frame from what disk database does. As the frame is specially designed for main memory, the MMDB gets a much higher(10 times more) performance in dealing data than what disk database does. Though MMDB has a good performance, it's not enough when facing huge of data dealing. With the appearance of SMP, one of the methods to enhancing the MMDB performance is to implement parallel query optimization.The parallel query optimization has face several challenges: parallel frame research and how to program on that frame, optimization of process issues in parallel, concurrency control, determine the extent of parallization and policy issues of concurrent execution. This paper puts forward one by one solution. We decide to choose OpenMP to implement parallel optimization after we researched the frame of MMDB and compared OpenMP and MPI. By way of comparison of various parallel query, including inter-query parallelism, inter-operation parallelism and inner-operation parallelism, finally select the inner-operation parallelism as a primary means of parallel query optimization. By analyzing the query process, we offer a parallel query optimization process and give a detail parallel processing strategy for index, operations and result set processing in the parallel query process. For concurrency control problem, we use an improved spin lock which has a extend scope of application and give a theoretical analysis of error rates. For the parallelism problem, this paper presents a formula to determine the degree of parallelism, and simplifies the parallelization design. For concurrent execution strategies, this paper proposes an improved T-tree index which is suitable for parallel query. For aggregate functions, analyzes their parallel conditions and parallel processing procedure. For operation parallelism, we research technology of division of the data, and implement the join, sort and other commonly used time-consuming query operation in parallel optimization, and parallel process the result set by using OpenMP.Finally, we verified the effectiveness of the parallel optimization by experiment and simulation and test the performance of improved spin lock operations, the performance of parallel sort operations and the performance of parallel processing result set. This paper shows the deficiencies of the thesis and proposes the future research directions in the end.