FNN-Cloud: A Hybrid Fuzzy-Neural Framework for Adaptive Resource Isolation in Multi-Tenant Cloud Environments

Abstract

This paper proposes a dynamic resource isolation framework FNN-Cloud based on fuzzy neural network (FNN), which aims to solve the limitations of static policies and the lack of ability to handle uncertain demands in cloud computing environments. FNN-Cloud is designed for multi-tenant scenarios. It uses fuzzy logic to quantify uncertain resource demands and dynamically adjusts isolation thresholds through neural networks to optimize resource utilization and maintain service level agreement (SLA) compliance. In terms of computational methods, the framework uses a double hidden layer back propagation (BP) neural network combined with an adaptive moment estimation (Adam) optimizer and a dynamic loss function (SLA violation loss + resource utilization loss) for online learning. At the same time, it uses triangular membership functions to fuzzify key indicators such as CPU utilization and memory pressure, and uses a 3×3 fuzzy rule base to handle multi-dimensional resource coupling relationships. In terms of experiments, 8 physical nodes are deployed on the OpenStack test platform to simulate three typical workloads: Web services, data analysis, and mixed workloads, and compared with static thresholds, long short-term memory networks (LSTM), and deep Q networks (DQN). Test data shows that FNN-Cloud outperforms the baseline model in CPU usage (28.3%--34.7%), memory usage (31.5%--37.2%), and SLA violation rate (2.1%--4.5%), while reducing P99 latency by 62.3% and controlling the policy response time within 51.4 milliseconds. The system demonstrates efficient and robust dynamic isolation capabilities through a fuzzy priority arbitration mechanism and a neural prediction-driven pre-isolation strategy, providing a reproducible intelligent optimization solution for cloud computing resource management.