Design and Implementation of Machine Learning-Based Fault Prediction System in Cloud Infrastructure

Abstract

The method for ensuring availability in an existing cloud environment is primarily a metric-based fault detection method. However, the existing fault detection method makes it difficult to configure the environment as the cloud size increases and becomes more complex, and it is necessary to accurately understand the metric in order to use the metric accurately. Furthermore, additional changes are required whenever the monitoring environment changes. In order to solve these problems, various fault detection and prediction methods based on machine learning have recently been proposed. The machine learning-based fault detection and recovery model most commonly proposed in the cloud environment is a supervised machine learning method that learns data relating to fault situations and, based on this data, detects faults. However, there is a limit to fault learning because it is difficult to obtain all of the fault situation data necessary to learn all of the fault situations that occur in a large-scale cloud environment. In addition, it is difficult to detect a fault when a fault that differs from the learned fault pattern occurs. Furthermore, it is necessary to discuss the automatic recovery architecture leading to the fault recovery procedure based on the fault detection results. Therefore, in this paper, we designed and implemented a whole system that predicts faults by detecting fault situations using the anomaly detection method.