As the capabilities and component count of systems increase, the mean time between failures decreases. Typically, applications tolerate failures with checkpoint-restart to a PFS. While simple, this approach can suffer from contention for PFS resources. Multilevel checkpointing is successful on many machines but insufficient for extreme-scale systems, which have orders of magnitude larger memory sizes and failure rates.

To solve the problem, we combined the benefits of asynchronous and multilevel checkpointing, and modeled the system. Our experiments showed that our system can improve efficiency by 1.1x to 2.0x on future machines. Additionally, applications using our checkpointing system can achieve high efficiency even when using a PFS with lower bandwidth.