Autonomous resource provision in virtual data centers

In recent years the advances in cloud computing and virtualization have created the need for autonomic resource provision. The correct provisioning of resources is a difficult task due to variations and uncertainty in workload demands. Most data center workload demands are very spiky in nature and often vary significantly during the course of a single day. Because the resource availability in a data center is generally unpredictable due to the shared feature of the cloud resources and because of the stochastic nature of the workload, severe service level agreement (SLA) violations may occur frequently. To overcome this problem, researcher's attention is diverted towards developing dynamic resource management strategies. In this paper, an autonomic resource controller is proposed that dynamically controls the resource allocation for data center's virtual containers. The controller has two parts: A resource modeler that models the non-linearity of the system by employing different Machine Learning techniques allowing the datacenter to allocate the appropriate resources and a resource fuzzy tuner that dynamically tunes the allocated resources using fuzzy logic to sustain the desired performance taking into consideration the enforcing of service differentiation among clients. Experimental results on a real data center dataset showed that the proposed resource controller can predict future resource needs while still sustaining performance goals stated in the SLA. Also, using the bagging and the boosting techniques along with model tree classifiers was demonstrated to be promising in terms of accuracy and performance.