OptiPlace: Designing Cloud Management with Flexible Power Models through Constraint Programing

The evolution of data centers infrastructures requests clever management systems, as the size of those data-centers can reach several hundreds of physical servers to run each tenths of Virtual Machines. The power consumption of those centers, estimated to 1.3 percent of the world's power consumption in 2011, can be managed using specific policies presented in the literature. One specific data center manager is Entropy, based on the Constraint Programming solver Choco written in Java. We extended Entropy to give it the support of external models, named views. Specifically, we developed a power view, based on generic servers' models, to define and reduce the power consumption of the data center's physical servers. It also relies on generic power models which can handle other resources than memory and CPU. We integrated heuristics and constraints in this view to solve data center reconfiguration problems with power-related constraints and objectives. On the contrary of what is proposed in the literature, this view can be integrated with other models in order to combine their heuristics and user constraints. In this paper, we present the extension we performed in Entropy, the power view we developed, the heuristics and constraints it integrates and we evaluate the power-reduction performances on several reconfiguration problems using our heuristics.