A New Parallel Recomputing Code Design Methodology for Fault-Tolerant Parallel Algorithm

As the size of large-scale computer systems increases, their mean-time-between-failures are becoming significantly shorter than the execution time of many current scientific applications. To complete the execution of scientific applications, they must tolerate hardware failures. Fault-tolerant Parallel Algorithm (FTPA) is an application-level fault-tolerant approach for large-scale scientific applications, and it can achieve fast self-recovery through parallel recomputing. In this paper, first we propose a new parallel recomputing code design methodology, and the parallel recomputing code designed by the methodology can achieve a high efficiency of parallel recomputing. Second, the parallel recomputing code design methodology is automated by exploring the use of compiler technology. Finally, we evaluate the performance of our approach with two kernels of NAS Parallel Benchmarks on a cluster system with 512 CPUs. The experimental results show that the parallel recomputing code generated by our approach has a higher efficiency of parallel recomputing than the code generated by loop parallelization.