Energy cost minimisation of geographically distributed data centres

In this paper we present a Mixed Integer Programming-based (MIP) approach to optimise the workload allocation of geographically distributed Data Centres (DCs) in the face of dynamic DC performances and electricity prices. We reduce the overall electricity cost for running a DC set over an operating horizon by finding a good compromise between: The number of migrations subject to the sovereignty of data, the loads of the servers in DCs and the energy cost reduction possible by following the DCs with best performance and energy efficiencies over time. To model the DC performance we use Power Usage Effectiveness (PUE), with a devoted function per DC dependent on the current outside temperature. We discuss the multiple dimensions of the problem, present a mathematical formulation for it and provide empirical evaluation to claim the improvement on the electricity cost achieved.