Performance Implications of Periodic Checkpointing on Large-Scale Cluster Systems