The rise of multi-tenant architectures in cloud computing has introduced complex challenges in maintaining tenant isolation, crucial for securing sensitive data and ensuring optimal performance. This study explores the application of machine learning (ML) techniques—specifically anomaly detection, predictive resource allocation, and dynamic isolation via reinforcement learning—in enhancing tenant isolation within these architectures. By evaluating models such as autoencoders, recurrent neural networks (RNN), and deep Q-networks (DQN), we assess the impact on security metrics, resource efficiency, and overall system latency and throughput. The findings reveal that ML-powered isolation not only improves threat detection and resource management but also reduces latency and enhances throughput, outperforming traditional isolation methods. However, the increased computational overhead of ML models and susceptibility to adversarial attacks pose challenges that warrant further investigation. This study underscores the potential of ML to balance security and performance demands in multi-tenant systems, offering a scalable solution for future cloud environments.
Jain et al. (Thu,) studied this question.