Data Stream Mining Based-Outlier Prediction for Cloud Computing

The cloud computing is the dream of computing used as utility that became true. It is currently emerging as a hot topic due to the important services it provides. Ensuring high quality services is a challenging task especially with the considerable increase of the user's requests coming continuously in real time to the data center servers and consuming its resources. Abnormal users requests may contribute to the system failure. Thus, it's crucial to detect these abnormalities for further analysis and prediction. To do that, we propose the use of the outlier detection techniques in the context of the data stream mining due to the similarity between the nature of the data streams and the users requests which require analysis and mining in real time. The main contribution of this paper consists of: first, the formulation of the users requests as well as the server state as a stream of data. This data is generated from CSG(+) a cloud stream generator that we extended from CSG [1]. Second, the creation of a framework for the detection of the abnormal users requests in terms of the CPU and memory by using AnyOut and MCOD algoithms implemented within MOA (Massive Online Analysis) (http://moa.cms.waikato.ac.nz/) framework. Third, the comparison between them in this context.