Enhancing Software-Defined Networking With Dynamic Load Balancing and Fault Tolerance Using a Q-Learning Approach

Abstract

The Software-Defined Networking (SDN) paradigm represents a fundamental shift in networking by decoupling the control plane from the data plane in network devices. This architectural change offers numerous advantages, including network programmability and centralized management capabilities, which improve scalability and efficiency compared to conventional network architectures. However, the dynamic nature of network traffic presents overload challenges, both temporally and spatially, especially in multi-controller SDN settings. To address these challenges, this paper presents an approach leveraging network traffic patterns for dynamic load balancing. The proposed framework optimizes migration strategies to reduce costs and enhance in-packet request-response rates. By exploiting load ratio variance across controllers, the architecture identifies optimal migration triplets, encompassing migration-in and migration-out domains by selecting a subset of switches. The architecture utilizes online Q-learning technology to achieve optimal controller load balancing while minimizing associated expenses. The proposed approach ensures stability and scalability by imposing limits to maintain maximum efficiency and reduce migration conflicts. It iteratively converges to an optimal policy through a comprehensive set of simulations performed on switches under a wide range of load distribution situations. These results highlight the effectiveness and adaptability of the proposed methodology in addressing the intricacies present in dynamic network settings, encouraging further progress in the field of SDN technologies and their real-world applications.