Mohammad Shuab Siddique

Debugging Billion-Core AI Clusters: Distributed Crash Diagnosis for Heterogeneous Accelerator Systems

Correlated kernel panics are a common distributed system reliability issue because modern AI systems involve billions of logical processing cores on a variety of NVIDIA, TPU, AMD, and specialized accelerator platforms. This study examined Cluster Crash Diagnosis (CCD) as an architecture that uses Linux kdump technology...