Log Mining-based Online Failure Prediction in Client-Server Architecture

Abstract

Client-server model is a widely used distributed system architecture that portions out the workload among the clients. Despite its advantages as a stable and secure architecture, it requires the server to have full control over the clients in order to ensure the continuity of the workflow. A seem-to-be minor fault in a single client might cause a considerable loss of time and resources. Hence, systematic monitoring of clients is becoming a must for the server. At this point, log mining arises as one of the most convenient ways to keep track of various resources. Event logs contain summarized information about the systems, thus, they become quite useful in terms of forecasting failures beforehand. In this study, we propose an online failure prediction method for client-server architecture based on log mining. The method employs CNN (Convolutional Neural Networks) as the deep learning algorithm, with the intent of reducing time and resource consumption. Test results demonstrate that the method provides a reliable prediction mechanism that achieves a TPR (True Positive Rate) of 99,41%. © 2023 Elsevier B.V., All rights reserved.