Study On .NET Framework Based Distributed Parallel Computing System

In modern scientific research and engineering, there are many computation intensive problems so that high performance computers are needed. Parallel computers are too expensive for most of the circumstances. Because of the flexible structure, high cost-performance ratio and easiness of buildup, computer clusters based on common PC and workstations are accepted by more and more people. The appearance and application of object-oriented technology make parallel computing and distributed computing more active. According to the research and analysis of current methods of parallel computing and key techniques for implementing parallel computing in a distributed way, we developed some algorithms of load-balance, fault-tolerance and scalability, which are adapted to be used in building distributed systems using distributed object technology. We further implemented a distributed parallel computing system based on .NET framework. Key contributions of this thesis are as follows: (1) Analyzed the characteristics and shortcomings of current programming methods of distributed parallel computing. (2) Developed a flexible load-balance algorithm-Mixed Load-Balance Algorithm, a high performance fault-tolerance algorithm-DOAC Algorithm and a model to solve the scalability problem of building distributed parallel computing systems using distributed object technology. Experiments showed that the use of these algorithms and model makes the system better in load-balance, fault-tolerance and scalability. (3) Based on the algorithms and model mentioned in the (2), we implemented a distributed parallel computing system based on .NET framework. And the system was tested using a test case, so that the feasibility, stability and efficiency of the system were proved.