An adaptive and distributed intrusion detection scheme for cloud computing

Abstract

Cloud computing has enormous potentials but still suffers from numerous security issues. Hence, there is a need to safeguard the cloud resources to ensure the security of clients’ data in the cloud. Existing cloud Intrusion Detection System (IDS) suffers from poor detection accuracy due to the dynamic nature of cloud as well as frequent Virtual Machine (VM) migration causing network traffic pattern to undergo changes. This necessitates an adaptive IDS capable of coping with the dynamic network traffic pattern. Therefore, the research developed an adaptive cloud intrusion detection scheme that uses Binary Segmentation change point detection algorithm to track the changes in the normal profile of cloud network traffic and updates the IDS Reference Model when change is detected. Besides, the research addressed the issue of poor detection accuracy due to insignificant features and coordinated attacks such as Distributed Denial of Service (DDoS). The insignificant feature was addressed using feature selection while coordinated attack was addressed using distributed IDS. Ant Colony Optimization and correlation based feature selection were used for feature selection. Meanwhile, distributed Stochastic Gradient Decent and Support Vector Machine (SGD-SVM) were used for the distributed IDS. The distributed IDS comprised detection units and aggregation unit. The detection units detected the attacks using distributed SGD-SVM to create Local Reference Model (LRM) on various computer nodes. Then, the LRM was sent to aggregation units to create a Global Reference Model. This Adaptive and Distributed scheme was evaluated using two datasets: a simulated datasets collected using Virtual Machine Ware (VMWare) hypervisor and Network Security Laboratory-Knowledge Discovery Database (NSLKDD) benchmark intrusion detection datasets. To ensure that the scheme can cope with the dynamic nature of VM migration in cloud, performance evaluation was performed before and during the VM migration scenario. The evaluation results of the adaptive and distributed scheme on simulated datasets showed that before VM migration, an overall classification accuracy of 99.4% was achieved by the scheme while a related scheme achieved an accuracy of 83.4%. During VM migration scenario, classification accuracy of 99.1% was achieved by the scheme while the related scheme achieved an accuracy of 85%. The scheme achieved an accuracy of 99.6% when it was applied to NSL-KDD dataset while the related scheme achieved an accuracy of 83%. The performance comparisons with a related scheme showed that the developed adaptive and distributed scheme achieved superior performance.