Predictive Modeling of Performance Variability in HPC Applications

Performance variability is common in HPC applications, due to factors such as job scheduling, resource contention, and network traffic. We propose a machine learning-based model to predict performance variability in high performance computing applications. Our model integrates classification and regression techniques to identify jobs with high variability and accurately predict their performance fluctuations. The model incorporates job-specific and system-wide factors such as network counters to enhance prediction accuracy.