A Deep Learning Based System for Traffic Engineering in Software Defined Networks

Özet

Traffic engineering is essential for network management, particularly in today's large networks carrying massive amounts of data. Traffic engineering aims to increase the network's efficiency and reliability through intelligent allocation of resources. In this paper, we propose a deep learning-based traffic engineering system in Software-Defined Networks to improve bandwidth allocation among various applications. The proposed system conducts traffic classification based on deep neural network and one dimensional convolution neural network models. It aims to improve the Quality of Service by identifying flows from various applications and distributing the identified flow to multiple queues where each queue has a different priority. Then, it applies traffic shaping in order to manage network bandwidth and the volume of incoming traffic. To increase the network's performance and avoid traffic congestion, we implement a technique that considers the port capacity to accomplish general load balancing. We solved the issue of imbalanced dataset by implementing an oversampling technique called Synthetic Minority Over-Sampling Technique. The performance of DNN and 1-D CNN have been compared and evaluated with some of machine learning models, such as KNN, SVM, DT, and RF. The results showed 1-D CNN and DNN are able to achieve more than %88 accuracy of traffic captured in 5s and 10s timeout, while KNN and RF are able to achieve more than %98 accuracy of traffic captured in 15s and 30s timeout, and the evaluation of the overall system showed applying traffic shaping to the identified flow increases the network's performance and bandwidth availability.