Deep Learning-Assisted Traffic Prediction in Hybrid SDN/OSPF Backbone Networks
Richard Etengu, Saw Chin Tan, Teong Che Chuah, Jaime Galán–Jiménez · NOMS 2022-2022 IEEE/IFIP Network Operations and Management Symposium · 2022
Deploying a real-world software defined network (SDN) requires instantaneous link traffic information. This has necessitated for the need of accurate real-time data analytics and traffic matrix (TM) prediction methods. So far, several frameworks have been developed to enable analysis and generation of valuable information from huge volumes of partial and noisy data. But, owing to the linear nature of network design, generally typified by manual control plane forwarding design, current frameworks are incapable of performing accurate traffic prediction over datasets in modern non-recurrent large-sized networks. To address this issue, deep learning (DL) methods have recently been proposed as a possible solution. But, deciding the most appropriate DL models to be employed for accurate TM prediction is still a challenge. This paper proposes an improved DL framework that utilizes different dimensionality feature reduction techniques to perform short-term TM prediction in SDN networks. The two dimensionality reduction techniques required to perform feature reduction for the DL model are correlation component analysis (CCA) and principal component analysis (PCA). Investigational results show that the proposed method can achieve more accurate forecast of link traffic in comparison to the traditional baseline machine learning frameworks.