The Analysis of Checkpoint Strategies for Large-Scale CFD Simulation in HPC System

With the development of the electronic technology, the processors count in a supercomputer reaches million scale, making the fault problem becomes a fundamental issue for massive parallel CFD simulation. Checkpoint/Rollback technology is a widely used fault tolerant method, and has a obvious affect for massive parallel application. In this paper, we explore the checkpoint method for the CFD simulation with the CFD simulation features, and analysis the two checkpoint strategies: fine granularity checkpoint and coarse checkpoint. We analysis the checkpoint intervals and the volume of the backup data, and their impact on the FT overhead through model. Experimental results on the Tianhe-2 supercomputer demonstrate that coarse checkpoint can achieve a much better FT effect for the CFD simulation.