Enabling Coordinated Checkpointing for Distributed HPC Applications

Explore a conference talk on enabling coordinated checkpointing for distributed HPC applications. Discover a new coordination mechanism for Checkpoint/Restore in User-Space (CRIU) and its integration with the Kubelet Checkpoint API to facilitate checkpoint and restore operations for distributed applications. Delve into the challenges of Pod preemption and handling failures in long-running workloads. Learn about potential extensions to the Kubernetes scheduler that utilize container checkpoints to roll back distributed applications to a globally consistent state. Gain insights into the latest advancements in container checkpointing across multiple Pods and cluster nodes, addressing the complexities of coordinating checkpoint/restore operations in distributed environments.