Performance Prediction and Anomaly Detection Using Deep Learning

This technical report outlines the methodologies employed by ProphetStor Data Services, Inc. for performance prediction and anomaly detection in data center operations using deep learning techniques. The report begins by discussing the increasing complexity of IT infrastructure and the role of artificial intelligence in enhancing data center management. It details the use of Long-Short Term Memory (LSTM) models to predict performance metrics based on historical data, specifically focusing on disk performance metrics. The report also describes the process of anomaly detection, highlighting how deviations between observed and predicted values can indicate performance issues. A novel approach is presented to minimize false alarms by comparing observed values against historical data over specific time intervals. The report concludes with an explanation of the advantages of deep learning over traditional algorithms for accurately modeling dynamic performance metrics and reducing false positives in anomaly detection.