| Parallel computing plays important roles and is considered as a standard tosolve problems in many applicative areas as engineering design, sciencecomputing and business. With the development of high-speed networktechnology, parallel computing has been extended into the network—networkparallel computing, which is characterized as cluster computing based on LANand Grid computing based on Internet. Cluster computing has become a hotresearch topic and the main trend of the high-performance computer system,because it provides computing power in a flexible and very low-cost way.PC cluster in LAN that based on parallel computing environment ofmessage passing interface have large potential abilities in network parallelcomputing. But now, there're many questions of parallel computing need to besolved by further research. How to enhance the abilities in performance andtask scheduling strategy and load balancing algorithm of parallel computingbecomes major focus for researchers now.This paper focused on the following aspects:Firstly, general theory of parallel programming environment was reviewed.Through analyzing the factors impacting on the performance of parallelcomputing, I proposed that the performance ability of parallel computing canbe enhanced by reducing the communication costs and utilizing feasible loadbalancing algorithm.Secondly, focus on the large communication overhead of matrix transportin parallel computing, aimed at reducing the communication overhead, a novelmulti-dimensioned matrix transport protocol was proposed.Thirdly, a new hybrid scheduling algorithm that combined with looprotation and weighted round dynamic load balancing algorithm was brought outto reduce the parallel computing time while the load was kept balanced.Finally, the experiment platform was construct based on LAN, and thedesigned experiments were tested on it by using the MPI-2 functions. And theexperiment results were analyzed. |
PDF Full Text Request