Root Cause Analysis Framework for Intelligent Computing Center

The rapid development of AI technology has greatly increased the demand for computational power, leading many companies to deploy large-scale intelligent computing centers. However, the larger cluster sizes result in more frequent failures, making root cause analysis more challenging. To address these issues, we propose ASeqRCA, a novel framework for root cause analysis in intelligent computing center. We simplify the complex network structure of intelligent computing clusters and model the relationships between alarms in sequences by frequent-pattern mining and causal learning techniques. With an alarm causal graph, we apply depth-first search to identify potential alarm propagation paths. We illustrate the effectiveness of the proposed method through a real-world AI model training failure case.