earticle

영어

Anomaly detection under Cloud computing environment plays an important role in detecting anomalous virtual machines (VMs) before real failures occur. In order to accurately characterize the new trend of VMs' performance, new samples are collected, detected, and selectively added into the training sample set. The newly added samples are used for updating the detection model, so as to improve detection accuracy. However, increasing number of training samples causes both much storage space and CPU time. To overcome this challenge, this article proposes an anomaly detection algorithm based on online learning Lagrangian SVM (termed OLLSVM) for detecting anomalous VMs. Online learning includes incremental learning and decremental learning. Single-sample and batch incremental learning algorithms are designed to update the detection model when adding a single sample or a set of samples. Similarly, single-sample and batch decremental learning algorithms are designed for deleting a single sample or a set of samples. The strategies for selecting sample(s) to be added or deleted are also designed. This article conducts experiments on Cloud datasets and KDD Cup datasets. The experimental results show that, compared with traditional Lagrangian SVM (LSVM) which retrains the detection model each time when adding or deleting sample(s), OLLSVM achieves almost similar high detection accuracy but much higher time efficiency.