Enhancing Video Transmission with Machine Learning based Routing in Software-Defined Networks

Overview

• This study uses Software-Defined Networks (SDN) to improve network performance.
• SDN provides a flexible and programmable network structure but requires further enhancements.
• The Floodlight controller used in this study has limited routing information, which is insufficient for video transmission scenarios.
• An AI-based routing algorithm is proposed to optimize video transmission between a server and client in an NSFNET topology.

Plain English Explanation

In traditional computer networks, the devices that route and forward data often have fixed, inflexible configurations. Software-Defined Networks (SDN) offer a more flexible and programmable approach, where a centralized controller can dynamically adjust the network's behavior. This study investigates using SDN to improve network performance, particularly for video streaming applications.

The researchers used a Floodlight SDN controller, which relies on simple metrics like hop count to make routing decisions. However, this approach is not sufficient for video transmission, where factors like latency, throughput, and packet loss are critical. To address this, the researchers developed an artificial intelligence (AI)-based routing algorithm to optimize the video transmission path between a server and client.

The researchers set up a simulated network environment using the Mininet tool, with a client, server, and 14 switches. They generated realistic network traffic using the iperf3 tool and transmitted video streams using FFmpeg. By analyzing 49 different network performance metrics, they were able to measure the quality of the video transmission under both congested and uncongested conditions.

Since there was no existing dataset suitable for this scenario, the researchers created a new dataset of 876 records from the simulated video traffic. They then applied various machine learning techniques, such as K-Nearest Neighbors (KNN), Random Forest, and Support Vector Machines (SVM), to predict network congestion levels and inform the AI-based routing algorithm.

Technical Explanation

The study leverages the centralized, flexible, and programmable nature of Software-Defined Networks (SDN) to address the challenges present in traditional networks. While SDN effectively addresses many of these challenges, the researchers identify the need for further enhancements to achieve a more optimized network architecture.

The Floodlight SDN controller used in this study employs simple metrics, such as hop count, to make routing decisions. However, in scenarios involving video transmission, this approach provides limited information and is insufficient for achieving the necessary optimization. To address this, the researchers propose an artificial intelligence (AI)-based routing algorithm to optimize the video transmission path between the server and client in an NSFNET topology.

The researchers designed the network topology in the Mininet simulation environment, which includes a client, a server, and 14 switches. To create a realistic network environment, the researchers added different receivers and generated TCP traffic between these receivers using the iperf3 tool. In three scenarios, video streaming was performed using the FFmpeg tool, and 49 path metrics, such as Round-Trip Time (RTT), throughput, and packet loss, were recorded.

Due to the lack of a suitable dataset in the literature for the proposed network environment, the researchers created a new dataset consisting of 876 records from the continuously transmitted video traffic. The dataset includes both low and high traffic levels, and the researchers applied various machine learning techniques, including KNN, Random Forest, SVM, AdaBoost, Logistic Regression, and XGBoost, to the dataset to predict the traffic levels based on the recorded features.

Critical Analysis

The researchers acknowledge the limitations of the Floodlight SDN controller's routing metrics, which are insufficient for optimizing video transmission. The proposed AI-based routing algorithm aims to address this by considering a more comprehensive set of performance metrics. However, the paper does not provide details on the specific algorithm or its implementation, making it difficult to assess its effectiveness.

Additionally, the creation of a new dataset for this scenario is a significant contribution, as the lack of suitable datasets is a common challenge in network research. However, the size of the dataset (876 records) may be relatively small, and the researchers do not provide information about the data collection process or the potential biases or limitations of the dataset.

Furthermore, the paper does not discuss the computational complexity or the training time required for the various machine learning techniques applied to the dataset. This information would be valuable for evaluating the practical feasibility of deploying the proposed solution in real-world scenarios.

Conclusion

This study demonstrates the potential of using Software-Defined Networks (SDN) and artificial intelligence (AI) to optimize network performance, particularly for video transmission scenarios. The researchers have identified the limitations of traditional routing metrics and proposed an AI-based approach to address these limitations.

The creation of a new dataset for this scenario is a valuable contribution to the field, as it can enable further research and development in this area. However, the lack of details on the proposed AI algorithm and the limited size of the dataset suggest that further work is needed to fully validate and deploy the proposed solution in real-world networks.