Consideration about Fast Checkpointing Mechanism for High-reliable HPC Cluster System

摘要

Cluster systems are getting widely used because of good performance / cost ratio. However, little attention has been paid for their reliability so far. As the number of commodity components in a cluster system gets increased, it is indispensable to support reliability by system software. We propose a hierarchical checkpointing in this paper. We explain its mechanism, and show preliminary experimantal results. In the experiment, the proposed mechanism is prototyped by modifying SCore cluster system, which is a parallel programming environment with checkpoint mechanism and is open to public.