| Data-intensive applications, most of which are with strong data dependencies, are widespread in scientific computing area such as bioinformatics, financial management, production scheduling, engineering and so on. Problems which can be decomposed into repeated sub problems and are with optimal substructure can be solved efficiently by the dynamic programming algorithm. However, due to the sharp increase in the size of data, the calculation time of the traditional serial program becomes unacceptable, even implemented using the efficient dynamic programming algorithm. An effective method is to design the parallel program to solve the strong data dependence applications. However, in consideration of the complexity of and easy error-prone of parallel programming, as well as the strong dependence, parallelizing the strong data dependence applications is very difficult.The DAG data driven model is proposed to simplify the parallelization of dynamic programming. This model allows users to develop parallel programs like serial programming, and the runtime system parallelizes the program and maintains scheduling process, so developers can focus their attention in applications and algorithms. However, there are still many deficiencies in the model.This paper firstly designs the parallel programming model for strong data dependence applications by studying the DAG data driven model and improving its designs focus on some imperfections. This model solves the problem in accessing shared data during main part and improves the scalability, portability and configurability of the structure, especially the scalability of DAG patterns component to more applications. Then, a system named EasyPDPv3 has been implemented on shared memory environment based on the parallel model for data intensive applications. This system is a practical tool for parallel program developers. Finally, a testing for the performance of EasyPDPv3 and its extensions on some parameters has been carried out. |
PDF Full Text Request