Research On The Fast Parallel Recomputing For Parallel Digital Terrain Analysis

With the rapid increase of spatial data resolution and data scale, traditional serial computation method cannot meet the needs of high performance applications and other aspects of science and production. The emergence of parallel computing can effectively solve the problems which serial computing cannot deal with. At present, the combination of parallel computing and digital terrain analysis has become one of the research hotspots in Geographic Information System (GIS). On the one hand, the development of parallel computing provides powerful data computation ability for the users and improves the utilization rate of resources. On the other hand, the application of parallel computing has greatly promoted the development of geoscience research and engineering practice. However, most research has focused on the parallel terrain analysis algorithm rather than the fault tolerance of digital terrain analysis. According to the features of parallel digital terrain analysis, how to introduce appropriate fault tolerance technique to improve the stability and reliability of the computation with DEM data has become a one of key problems and hot topics in high-performance parallel digital terrain analysis.Firstly, in view of data and computation intensive characteristics of parallel digital terrain analysis and the characteristics of parallel computing platform, the thesis proposes an error-detecting approach for fault tolerance recomputing oriented to parallel digital terrain analysis. The method adopts MP1 program model and OpenMP program model to detect the errors of the computation results with DEM data. Based on effective data partitioning strategy, the thesis implements the data computation and error detecting in parallel. It not only saves resources, but also reduces the overall time cost. At the same time, the thesis also puts forward an improved method to speed up the error detection by using the synchronization of data detecting and computing.Secondly, in order to ensure the correctness of computation results and reliability of the parallel digital terrain analysis, the thesis presents a new fault tolerance technology——Fast Parallel Recomputing (FPR). The fast parallel recomputing owns fast error recovery ability, when detecting data computing errors and all the idle processes will recompute the original data block to ensure the correctness of the final results. The thesis partitions the original data block into many logical data sub-blocks and stores the results of computation. When the computing result of a logical sub-block is found to have errors, the system immediately starts the recomputing procedure but permits the computation of the original data block to continue which can improve the effectiveness of the fault-tolerant parallel computing.Finally, the thesis implements the FPR method and the performance of the FPR is verified by the experiment results of the slope algorithm. It not only has advantages in the average time of the computation than the conventional fault tolerance technology, but also can solve the problem for error-detection and recovery in the parallel digital terrain analysis.