Recent years have witness the booming of the cloud computing, which provides customers with guaranteed services. Since any violation would inevitably lead to a degraded quality of service (QoS), anomalies detection has become a demanding task in cloud environments. To solve the above problems, the unsupervised clustering approaches were put forward for identifying those anomalies. However, they all failed to work out the anomalies detection from an evolutionary view. In this paper, we present a cloud anomalies detection framework called eCAD. Motivated by the evolutionary clustering, our eCAD employs an evolutionary algorithm with DBSCAN to detect cloud anomalies as time steps. Besides, we also propose an M-Nearest Neighbors (MNN) algorithm to conduct the inference for those induced anomalies. Our eCAD is evaluated on the top of VICCI platform, which is a federated cloud test-bed in IaaS level. As demonstrated in our experiment, our framework shows an advantage over its counterparts.

No abstract