An Efficient DDoS TCP Flood Attack Detection and Prevention System in a Cloud Environment

作者：Sahi, Aqeel^*; Lai, David; Li, Yan; Diykh, Mohammed

来源：IEEE Access, 2017, 5: 6036-6048.

DOI：10.1109/ACCESS.2017.2688460

• Classification
• cloud computing
• DDoS attacks
• LS-SVM

Although the number of cloud projects has dramatically increased over the last few years, ensuring the availability and security of project data, services, and resources is still a crucial and challenging research issue. Distributed denial of service (DDoS) attacks are the second most prevalent cybercrime attacks after information theft. DDoS TCP flood attacks can exhaust the cloud's resources, consume most of its bandwidth, and damage an entire cloud project within a short period of time. The timely detection and prevention of such attacks in cloud projects are therefore vital, especially for eHealth clouds. In this paper, we present a new classifier system for detecting and preventing DDoS TCP flood attacks (CS_DDoS) in public clouds. The proposed CS_DDoS system offers a solution to securing stored records by classifying the incoming packets and making a decision based on the classification results. During the detection phase, the CS_DDOS identifies and determines whether a packet is normal or originates from an attacker. During the prevention phase, packets, which are classified as malicious, will be denied to access the cloud service and the source IP will be blacklisted. The performance of the CS_DDoS system is compared using the different classifiers of the least squares support vector machine (LS-SVM), naive Bayes, K-nearest, and multilayer perceptron. The results show that CS_DDoS yields the best performance when the LS-SVM classifier is adopted. It can detect DDoS TCP flood attacks with about 97% accuracy and with a Kappa coefficient of 0.89 when under attack from a single source, and 94% accuracy with a Kappa coefficient of 0.9 when under attack from multiple attackers. Finally, the results are discussed in terms of accuracy and time complexity, and validated using a K-fold cross-validation model.

• 出版日期2017