On prediction to dynamically assign heterogeneous microprocessors to the minimum joint power state to achieve Ultra Low Power Cloud Computing

Cloud computing centers are designed to be scalable and to process large varieties of software applications. However, the total power required by cloud computing systems is high since excess processors must be available to service both on-demand applications, as well as existing processes. We describe novel concepts that can enable the introduction of Ultra Low Power Cloud Computing systems. Our approach involves using a variety of heterogeneous processors, each with different power and performance capabilities. By predicting the load and jointly allocating tasks to the processors and dynamically turning off reserve processors, we prove that power reductions of up to 60-80% can be achieved.