Smart Monitoring Implementation for Hosting Services Using Zabbix and Autoencoder Models

Abstract

Purpose: This study answers the increasing need for intelligent server monitoring systems capable of anomaly detection for computing infrastructure performance and reliability. Conventional static threshold-based monitoring systems, such as Zabbix, have limitations in identifying complex and dynamic abnormal patterns. To overcome this limitation, in this research study, the integration of an Autoencoder-based deep learning model with the Zabbix monitoring system is presented.

Methods: An experimental approach was followed, wherein important server performance metrics like CPU utilization, memory utilization, and network activity were collected every 15 minutes for 40 days. The Autoencoder was trained on normal operational data to identify frequent behavioral patterns and produce reconstruction errors as anomaly scores. Deviations were alerted as anomalies when these scores exceeded an adaptive threshold. The system was deployed in a virtualized environment, with real-time notifications sent via Telegram for rapid incident response.

Results: Experimental results show that the system can effectively detect different anomaly scores. Deviations were alerted as anomalies when these scores exceeded an adaptive threshold. The system was placed in a virtualized production setup, publishing real-time notifications via Telegram for prompt incident different anomaly types with high performance anomaly detection with MAE = 0.0084 and RMSE = 0.0198.

Novelty: The primary contribution of this study is the development of hybrid real-time anomaly detection framework that integrates deep learning within open-source monitoring tools to be more efficient, decrease the dependence on manual configuration, and enable computing infrastructures to be more reliable and efficient.