Request-Aware Fuzzy Load Balancing for Heterogeneous Computing Systems
Abstract
In modern heterogeneous computing systems, efficient load balancing remains one of the key challenges affecting system performance, response time, and resource utilization. Traditional load balancing algorithms such as Round Robin and Least Connection generally distribute requests without considering the computational complexity and priority of incoming tasks, which may lead to resource imbalance and performance degradation under dynamic workloads. To address this limitation, this study proposes a Request-Aware Fuzzy Load Balancing (RA-FLB) model based on a Mamdani-type Fuzzy Inference System (FIS). The proposed approach evaluates both request characteristics, including URL structure, payload size, header information, and computational weight, together with the real-time state of virtual machines such as CPU utilization and workload level. Based on fuzzy inference rules, the system dynamically selects the most appropriate server for each incoming request. In addition, a dynamic feedback mechanism continuously updates server states after task execution, enabling adaptive and real-time decision-making. The proposed model was implemented and evaluated in the CloudSim Plus simulation environment. Experimental results demonstrate that the RA-FLB approach improves response time, throughput, and load distribution efficiency compared with conventional algorithms. The proposed method provides a scalable and adaptive solution for intelligent resource allocation in cloud and distributed computing environments.