Approximate Dynamic Programming Based Data Center Resource Dynamic Scheduling for Energy Optimization

As the core part of modern IT infrastructure, data center consumes large amount of energy, which has become the main operational cost. In order to save energy consumption and reduce emission, it´s necessary to apply online dynamic scheduling of computational resources and physical resources for division of load, so as to cater for the need of time-variant and random service needs. The paper initiates the layered algorithm for scheduling of data-center resources and establishes energy-consumption models with tractable approximating computations for the data center and, on the basis of approximation dynamic programming method, establishes dynamic scheduling models of large-size heterogeneous resources and the algorithm for learning-based dynamic scheduling of resources. In order to evaluate the fidelity and efficiency of the models, the EnergyPlus and GreenCloud software are integrated into an analogue platform where simulation experiments are conducted and prove the efficiency of the model and the algorithm.