AIRSDN: AI based routing in software-defined networks for multimedia traffic transmission

Abstract

With the rapid increase in internet usage and the number of network-connected devices, network management and optimization have become increasingly challenging, particularly for high-bandwidth applications such as video streaming. The decentralized structure of traditional networks and the lack of standardization further complicate these challenges. Software Defined Networking (SDN) has emerged as a solution, enabling a more flexible and programmable architecture by centralizing network control. However, existing SDN controllers typically determine the optimal path based on simple metrics such as hop count and bandwidth, which can be insufficient in high-traffic scenarios. To overcome these limitations, this study proposes a novel artificial intelligence (AI)-based routing algorithm. Operating within the SDN framework, the proposed algorithm analyzes network traffic levels and dynamically selects the most efficient data transmission paths. The proposed algorithm is simulated in Mininet, a virtual network environment, using a network model inspired by real-world internet structures (NSFNET). Simulations are conducted under varying traffic conditions, with TCP (Transport Control Protocol) data and video transmission scenarios. Key performance metrics are observed, including round-trip time (RTT), throughput, packet loss, and video quality (measured using PSNR and SSIM). The machine learning model was trained using a custom dataset consisting of 876 records generated in the Mininet environment. Although the dataset size is sufficient for the simulation environment, caution should be exercised when generalizing the results to real-world network conditions. Future studies may aim to enhance the model's reliability by exploring data augmentation techniques and utilizing larger datasets that include real-world data. To classify traffic levels, machine learning models are trained, and the best-performing model (Logistic Regression) is integrated into the proposed routing algorithm. The results demonstrate that the proposed AI-based routing algorithm significantly improves network performance compared to both traditional hop-count-based and QoS-aware routing. Particularly in high-traffic scenarios, it achieves lower latency, higher throughput, and better video quality. Additionally, resource usage was analyzed on a Raspberry Pi 5 device, revealing stable RAM consumption (∼50 %) and fluctuating CPU utilization (10–90 %), indicating the feasibility of lightweight deployment with awareness of processing load. This study highlights the potential of AI-driven SDN frameworks for adaptive and efficient network traffic management in high-demand applications, offering a robust solution for dynamic routing.