Towards Transparent Checkpointing with AI-driven Code Generation
For HPC developers and resilience engineers, this work demonstrates that LLM-driven automation can reduce the expert effort required to add checkpoint/restart support to MPI applications.
The paper shows that a frontier LLM can automatically generate checkpoint/restart code for MPI scientific applications without human intervention, achieving working code in 50 minutes on average with negligible overhead on five of six applications and recovery efficiency comparable to human implementations.
Adding reliable checkpoint/restart support to an MPI scientific application is a time-consuming expert effort that requires deep knowledge of both the application and resilience. We ask whether a frontier large language model can perform this work end-to-end without human intervention. We assemble a benchmark suite of MPI applications spanning diverse domains and computation patterns, and drive an iterative code-generation loop for each application using Anthropic's Claude Opus 4.7 invoked through the OpenCode CLI. Across six scientific applications, the LLM generates working checkpoint/restart code in 50 minutes on average while consuming 3.4 M tokens per application. The generated code adds negligible overhead during normal failure-free execution on five of six applications and recovers from injected process failures with efficiency comparable to human-engineered checkpoint/restart implementations. These results suggest that automated end-to-end LLM-driven resilience engineering is technically viable today for a meaningful fraction of HPC applications.