Authors: Avinash Yeluri
Abstract: The fast embrace of multi-cloud computing has made it possible for organizations to gain scalability, flexibility, availability, and cost efficiency. Nevertheless, the monitoring of workloads in multi-cloud environments has become increasingly difficult because of fragmented visibility, privacy issues, increased communication overhead, and increasing telemetry data. Centralized monitoring methods involve sending information from distributed locations to a centralized location, resulting in higher costs and increased risks. The FL-IMME framework, introduced in this study, utilizes federated learning and is aimed at intelligent monitoring that preserves data locality. The framework combines telemetry gathering, generation of local intelligence, federated aggregation, anomaly detection, predictive monitoring, and adaptive decision-making mechanisms to ensure privacy-preserving and scalable monitoring capabilities. The novel approach involves training of local monitoring models in each cloud environment individually and exchanging encrypted model parameters via federated learning. Experiments were performed on a large multi-cloud observability data set and the performance was compared to CCFRL and AHFLP methods. The results revealed better performance of FL-IMME with anomaly detection accuracy of 99.0%, failure prediction accuracy of 99.1%, privacy preservation score of 99.4%, and system reliability of 99.5%. In addition, the proposed framework lowered the communication overhead and monitoring latency while increasing the efficiency of resources utilization.
DOI: https://zenodo.org/records/22957300
