Search

DDoS attacks, also known as distributed denial-of-service attacks, pose a significant risk to networks in the cloud. The attackers aim to flood the target system with an overwhelming amount of data and requests until it becomes completely overloaded and unable to function properly. These attacks are becoming smarter and more dangerous all the time. A low-rate DDoS attack is one such strategy that makes detection difficult. At the same time, cloud infrastructure is rapidly evolving. Container-based technology makes it possible for cloud computing to use resources efficiently and scale services in a flexible way. Existing methods for detecting DDoS attacks in cloud computing are insufficient when adversaries use low-rate DDoS attacks. A method is required that can not only identify the attack but also prevent it to some extent. A Low-Rate DDoS Attack Detection Framework (LRDADF) was proposed for this purpose when adversaries use low-rate DDoS attacks. A comprehensive approach is required because low-rate DDoS attacks are difficult to detect. In addition to employing deep learning methods to detect such attacks, we proposed a mathematical model to realize a mitigation strategy. As a result, we proposed a new algorithm called the Hybrid Approach for Low-Rate DDoS Detection (HA-LRDD). The algorithm employs an AI-enabled method comprised of deep convolutional neural networks (CNN) and a deep auto encoder. We defined another algorithm called Dynamic Low-Rate DDoS Mitigation (DLDM), which mitigates the impact of an attack once it has been identified. It also ensures that the attack is defeated and that the infrastructure continues to operate. A comprehensive simulation study revealed that the proposed framework is capable of detecting and mitigating low-rate DDoS attacks to ensure an acceptable level of service in cloud computing environments.

Distributed Denial-of-Service (DDoS) attack has been a serious threat to the availability feature of cloud computing. As traditional DDoS attacks are implemented using a huge volume of malicious traffic, the detection of such attacks becomes a naive task. To evade this detection, attackers are moving towards the Low-Rate DDoS (LRDDoS) attacks. The stealthy behavior of LRDDoS attack makes it difficult to get detected due to its low volume traffic. The existing frequency-domain approaches for LRDDoS detection are not feasible in terms of computational and storage requirements. This paper aims to propose a lightweight, accurate, and adaptive approach for the detection of LRDDoS attacks in frequency-domain. In this paper, the LRDDoS attack is detected by analyzing the power spectral distribution. The novelty of the proposed approach is to calculate the power spectral density using Fast Hartley Transform (FHT). The FHT processes real-valued input data, and has low computational and storage complexities. The approach is implemented on OpenStack cloud platform, and the aggregate network traffic (external and internal) is captured and analyzed. Experimental results show that the computational and storage complexities involved in FHT are lower than other transformation algorithms' complexities. Thus, the approach provides faster response with an average detection time of 60.16 s. The average true negative and true positive rates obtained by the proposed approach are 99.83% and 99.46% respectively, which are competitive. [ABSTRACT FROM AUTHOR]