ASYNCHRONOUS CHECKPOINT ACQUISITION IN PARALLEL COMPUTER CALCULATION OF ITERATION METHOD AND RESTORATION FROM THERE

摘要

PROBLEM TO BE SOLVED: To acquire a checkpoint when advancing computer calculation of an iteration method in parallel, and to efficiently utilize the acquired data upon restoration.SOLUTION: When acquiring the checkpoint in parallel calculation of repeating the iteration method such as time evolution calculation, calculation is not stopped independently in each node and the checkpoint is acquired in parallel with the calculation. Thus, the need of stopping the calculation during checkpoint acquisition time is eliminated, and the calculation and the checkpoint acquisition are simultaneously performed. When the calculation is not an I/O bottleneck, the checkpoint acquisition time is concealed and execution time is reduced. In the method, checkpoint data including values at different points of time during acquisition processing is acquired, and by limiting a use to convergence calculation of iteration method, coexistence of the values at the different points of time in the checkpoint data is allowed in the problem that a convergence destination is independent of an initial value.