International Journal of Computer Networks and Applications

Thangaraj Ethilu ¹, Abirami Sathappan ¹, Paul Rodrigues ²

Affiliations
1 Department of Computer Science and Engineering, Annamalai University, Chidambaram, Tamil Nadu, IN
2 Department of Computer Science and Engineering, DMI College of Engineering, Chennai, Tamil Nadu, IN

Abstract

Software Defined Networking (SDN) has increased a high-level attention in recent years, mainly because of its ability to address the cyber security challenges. Machine learning architectures were developed as the SDN system to detect the security threads; however, present techniques are limited with (i) higher computation time during malicious switch detection, (ii) reduced malicious switch detection rate (MSDR). This paper presents modified deep learning architecture based SDN system consist of two stages: (i) training stage, computes the external feature maps from both trusted and malicious network switches connected to the SDN controller, (ii) testing stage, classifying the trust and malicious switches connected with SDN controller. The feature maps are trained and classified with Modified LeNET Convolutional Neural Networks (CNN) architecture. The proposed methodology is simulated via network simulator under environmental constraint conditions. The results shows that the proposed methodology reduced the malicious switch detection computational time about a half as well as it increased the MSDR to about 6% compared to the conventional methodologies.

Keywords

SDN, Switch, Malicious, CNN, Feature Maps.

Full Text

   

Thangaraj Ethilu ¹, Abirami Sathappan ¹, Paul Rodrigues ²

Affiliations
1 Department of Computer Science and Engineering, Annamalai University, Chidambaram, Tamil Nadu, IN
2 Department of Computer Science and Engineering, King Khalid University, Abha, SA

Abstract

In recent times, Software Defined Networking (SDN) has developed widely to provide capable solutions for future internet services. As with the solutions, SDN brings us a hazardous rise in malicious threats. We investigated a sort of Distributed Denial of Services (DDoS) assault known as an internet services attack, which evaluates the influence of both traffic flow and throughput depletions in order to characterize the abnormalities. This sort of attack has a significant impact on the whole SDN. This paper introduces a deep learning method to improve the performance efficiency of the SDN by classifying the network switch into either a trusted switch or a malicious switch device. In this research, an attack detection methodology for Internet services utilizing Software Defined Networking (SDN) is proposed. The SDN controller may evaluate traffic flow, detect anomalies, and restrict both incoming and outgoing traffic as well as source nodes. The SDN considers a Convolutional Neural Network (CNN) based attack detection system that can identify malicious node. Kaggle datasets are used to test and train CNN and the features such as packet duration, packet count, byte count, accuracy for identifying the flow of trusted and malicious switches. According to the results, the CNN-based attack detection system can identify the attack with an accuracy of 89 percent. The comparison evaluation with the already proposed LeNet CNN of the feature classification proves that the flow is the trusted one and with the constant throughput with the help of the deep learning model.

Keywords

Software Defined Networking (SDN), Kaggle Dataset, Convolutional Neural Networks (CNN), Keras, Internet Service Attack, Malicious Switches, Malicious Node, Distributed Denial of Services.

Full Text