A Multi-Phase DRL-Driven SDN Migration Framework Addressing Budget, Legacy Service Compatibility, and Dynamic Traffic

Abstract

Software-Defined Networking (SDN) is a network architecture that offers enhanced flexibility, programmability, and more efficient traffic load management by decoupling the control plane from the data plane. However, complete migration to SDN is challenging for most organizations due to costs and operational complexities. Hybrid SDN has emerged as a practical incremental path where legacy and SDN-enabled nodes coexist, yet existing migration strategies typically address only individual challenges such as dynamic traffic patterns, legacy service compatibility, and budget constraints. This paper introduces SMART (SDN Migration Assisted by a Deep Reinforcement Learning (DRL) Technique), a comprehensive framework that simultaneously addresses dynamic traffic patterns, legacy service compatibility, and phased migration under budget constraints. By integrating a DRL model with a clustering algorithm, SMART determines the migration sequence to minimize link utilization and reduce the number of SDN-enabled nodes required for effective traffic load distribution under dynamic traffic patterns. Extensive evaluations on the Abilene and GEANT network topologies demonstrate that SMART outperforms three existing approaches, achieving most of the SDN benefits by migrating only 36% and 52% of legacy nodes, respectively. This approach can potentially lower migration costs by up to 64% while achieving network optimization objectives. These insights provide both a foundation for future research in network migration strategies and practical guidance for organizations planning cost-effective transitions from legacy to SDN-based architectures.