An SDN-Enabled Wireless HetNet Emulator to Analyze Effectiveness of Load Balancing Algorithms

Load balancing (LB) is crucial for wireless heterogeneous networks (HetNets), with one of the key challenges lying in the trade-off between optimality and computational complexity. To date, many studies have explored learning-based LB methods to reduce computational complexity. However, the impact of algorithm runtime on network performance is nontrivial, which has been widely neglected in current literature. Moreover, realistic network operations, such as packet forwarding, buffering, and routing, have also been overlooked, which would affect achievable throughput. To bridge this gap, we build a software-defined network (SDN)-enabled wireless HetNet emulator that considers the above factors, exemplified by an indoor hybrid light fidelity (LiFi) and wireless fidelity (WiFi) network (HLWNet). Specifically, the Mininet-based emulator integrates both wireless channel characteristics and end nodes’ movement features. In the built emulator, we implement and evaluate several LB methods, including game theory (GT), learning-based, and rule-based algorithms. Results show that traditional high-complexity algorithms (such as GT) exhibit a significant lag effect, which can degrade the network throughput by up to 60%. In contrast, the learning-based algorithm encounters a throughput reduction of less than 5%.