Authors:
M. S. Antony Vigil, M. Harish, M. Ripudaman, Sasi Dharan, Malini Premakumari William

Addresses:
Department of Computer Science and Engineering, SRM Institute of Science and Technology, Ramapuram, Chennai, Tamil Nadu, India. Department of Computer Systems Engineering, Mirage Software Inc., DBA Bourntec Solutions, Schaumburg, Illinois, United States of America.

Abstract:

Today, the seamless operation of servers is essential to the functioning of practically every aspect of our lives. An entire business can be thrown into disarray when it falls unexpectedly, leading to a loss of time, money, and even client trust.  We have devised a method for handling that that is more intelligent.  We can detect warning indicators at an early stage and provide IT teams with advanced notice before problems arise by utilising machine learning. We utilised a dataset that mimics server behaviour in the real world, which included measures such as CPU utilisation and disk activity, to conduct experiments on several algorithms. The algorithms we examined included Random Forest, Support Vector Machines, XGBoost, Neural Networks, and k-Nearest Neighbours, among others.  Random Forest distinguished itself from the other algorithms by achieving a remarkable accuracy of 99.98% in detecting early warning indicators of impending downtime. XG Boost came in a close second with an accuracy of 99.95%. With the use of Prometheus, a live data tracking system, and Grafana, a platform that provides simple-to-understand dashboards.  

Keywords: Server Downtime Prediction; Machine Learning; Random Forest; XG Boost and KNN; IT Infrastructure; Predictive Maintenance; Prometheus and Grafana; Support Vector Machines; Neural Networks.

Received on: 25/07/2024, Revised on: 15/10/2024, Accepted on: 12/11/2024, Published on: 03/03/2025

DOI: 10.69888/FTSCS.2025.000377

FMDB Transactions on Sustainable Computing Systems, 2025 Vol. 3 No. 1, Pages: 35-45

• Views : 75
• Downloads : 5

Download PDF