Performance models oriented to the dynamic resource provisioning in shared data centres

It is a challenge to quickly supply performance numbers online driving dynamic resource provisioning in shared centres in face of the complication of applications both in scale and architecture. In this paper, we provide a practical solution to the above problem by laying out a theoretical framework. In order to improve the representative characterization of workload, we classify the workload into classes and adopt the regression-based methodology to extract these parameters online. We constructed both effective open and closed queuing models to evaluate the correctness and the generality of our idea. In our experiments, we analyse the effectiveness of the regression method with different number of classes. The results for the performance evaluation in the open queuing network show that almost 100% tests show relative error less than 2%, so these performance indexes can be effectively used as the basis for autonomic resource provisioning.