Detecção de ataques de DDoS ao plano de controle de sdn utilizando aprendizado de máquina

Abstract

The Software Defined Networking (SDN) paradigm is considered promising for the innovation of computer networking technologies. The SDN architecture separates the data plane from the control plane, where the controller has an overall view of the network. Network security is a subject under constant discussion, as new forms of attacks with different objectives appear daily. SDN is no different, many Distributed Denial of Service (DDoS) attacks are performed against the SDN control plane, therefore, protection measures must be developed to detect malicious activities on the network. While SDN networks provide strong control over traffic, they also offer new problems and challenges as, for example, a DDoS attack against a controller has the potential to let the entire network inoperable. In order to identify malicious traffic in SDN, in this work, the input flows were analyzed and classified to detect DDoS attacks through machine learning techniques. In order to identify crucial characteristics in the monitoring of a SDN, datasets were created from the capture of legitimate and malicious traffic (DDoS) in SDN. These datasets were used in the construction of machine learning models which, in turn, were used to classify flows as legitimate or malicious. The traffic classification experiments were divided into two scenarios, one with variable traffic during the experiment time and another with unchanging traffic for each iteration. The results showed that the Naïve Bayes algorithm was more assertive in identifying attacks than the other algorithms (Gradient Boosting, Decision Tree and Support Vector Machine). To evaluate the results, the metrics accuracy, precision, recall and F-score were used.