Researches And Application Of Mapreduce Parallel Programming Model For Cloud Computing

The development of High performance computer promotes the production and development of High performance programming environment, especially for parallel programming model. On the one hand, because that the diversity of parallel computing computer’s architecture has brought a lot of difficulties to design of parallel program, how to simplify parallel program to improve designing efficiency has become a key issue. On the other hand, when facing cloud computing how to get valuable information in mass data by improving performance of parallel computing has also been a problem that desires to be solved urgently. Aiming at the two above-mentioned problems, Google had developed a new parallel programming model MapReduce, which could solve the problem of parallel computing simply and elegantly in high level of abstraction.Our research is Based on the study of parallel programming model and concrete parallel programming model, closely around cloud computing and its key technology MapReduce. Afer analysis and application of its implementation framework, implementation platform and execution process, The following achievements have been obtained:1. Aiming at a lack of quick and efficient rules to adjust and optimize performance when jobs are executed by Mapreduce, we propose several method to improve its performance. Firstly describe the problem, then give abstract resolution and do experiment at last. The comparatively experimental data is showed.2. According to the analysis of process executed by MapReduce, aiming that single-source shortest path algorithm is difficult to apply and improve search efficiency with the emergence and development of cloud computing. We designe and implemente a Parallel method of single-source shortest path algorithm based on MapReduce framework, and carry out research and experiments on Hadoop, test its parallel speedup, efficiency and scalability in detail. Results showed that the proposed algorithm can effectively search the single-source shortest path in the whole graph structure and verify its superiority of proposed algorithm.