Automated LSTM-Based Pipeline for HPC Data Center Node Risk Factor Prediction

View at publisher Summarise with VIRAT
Authors: Namratha H J, Burungale Swapnil Birudev, Pankaj Saini
Cited by: 0
References: 18

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

High-Performance Computing systems generate volumes of telemetry data continuously, and proactive node-level risk estimation is key for cluster reliability. This work proposes an automated LSTM-based predictive framework for node-level risk estimation in an HPC data center. The developed system receives multi-parameter telemetry input-power consumption, CPU temperature, and computational load-and computes the normalized risk factor and forecasts node behavior. A fully automated pipeline executes every day with no human intervention, performing timestamp correction, data merging, preprocessing, model training, and generating predictions. Each independent HPC node is modeled by a temporal LSTM network trained with six months of operational data. The suggested approach yields a low average MAE of 0.018 for 422 nodes, or 98.2% accuracy, displaying strong temporal pattern capturing and operational deviations. The proposed framework allows for proactive maintenance through early anomaly awareness with next-day (24-hour) risk predictions provided for all cluster nodes.

Fields of Study & Subjects

Reliability and Maintenance Optimization Cloud Computing and Resource Management Software System Performance and Reliability

Publication details

Journal / source
International Conference on Communication, Computing and Electronics Systems (ICCCES)
Pages
570–575
Publication date
21 January 2026
Language
EN