Probabilistic failure diagnosis in large-scale HPC applications with automaDeD

This article consists of a collection of slides from the author´s conference presentation. Some of the specific conclusions presented include: 1. The debugging approach, AutomaDeD, diagnoses failures in HPC applications at large scale; 2. Distributed debugging method is scalable (it takes fraction of a second with 32,000 MPI tasks); 3. Tests were performed with difficult-to-catch real-world bug and fault injections in HPC benchmarks. Possible future directions are noted.