A fast parallel re-computation with redundancy mechanism for parallel digital terrain analysis

According to many published literature, parallel computing is regarded as an efficient solution in digital terrain analysis (DTA) of geographic information system. The stable and credible services play an irreplaceable role in the high performance computing, especially when an error occurs in large-scale science computing. In this paper, a new approach for the parallel DTA considering the performance of fault-tolerance was proposed: fast parallel re-computation (FPR). FPR owns a fast self-recovery ability based on redundancy mechanisms compared to other fault-tolerant methods. Once some errors in application layers are detected, the data block having computation errors is further partitioned into several sub-blocks, which are re-computed by the surviving processes concurrently to improve the efficiency of failure recovery. The overlapping strategy of error detection and re-computation is presented through decomposing the data block into several logic sub-blocks. As a result, when an error of a logical sub-block of the data block is detected by a comparing thread the re-computing process immediately starts to correct the error. This strategy reduces the time of re-computation and error detection by overlapping them comparing the traditional re-computation method. The experiments show that the proposed FPR method can achieve better performance efficiency with fewer overhead.