Optimizing Checkpoint Intervals for Reduced Energy Use in Exascale Systems

In today's high performance computing (HPC) systems, the probability of applications experiencing failures has increased significantly with the increase in the number of system nodes. It is expected that exascale-sized systems are likely to operate with mean time between failures (MTBF) of as little as a few minutes, causing frequent interrupts in application execution as well as substantially greater energy costs in a system that will already consume large amounts of energy. State-of-the-art HPC resilience techniques proposed for use in these future systems complicate the energy problem further as the overhead associated with utilizing these techniques also further increases energy use. While work has been done that attempts to analyze and improve the energy use of systems utilizing resilience techniques, our work offers a new approach through the optimization of checkpoint interval lengths that allows a system designer the freedom to choose between intervals that optimize for application performance efficiency or energy use in both a traditional checkpoint and multilevel checkpoint approach to resilience. We create a set of equations able to optimize for either performance efficiency or energy use, demonstrate that distinct intervals exist when optimizing for either one metric or the other, and examine the sensitivity of this phenomena to changes in several system parameters and application characteristics.