Stochastic Programming for Cost Optimization in Geographically Distributed Internet Data Centers

The development of cloud computing has accelerated the worldwide growth of internet data centers (IDCs). While a large portion of the energy consumption generated by intense computation introduces greater operation expenditures to the IDC enterprises. To manage the overall costs and utilize resources to their fullest extent, this paper introduces the concept of spatio-temporal workload allocation among the geographically distributed IDCs within a cloud, with the guarantee of the workload completion time and the consideration of computing service delay penalties by introducing the cost of inconvenience. Apart from the effort of the workload migration, the spatio-temporal variance of the renewable energies in the data center microgrids (DMGs) is fully considered in this paper. What's more, as the power consumed by the IDCs are primarily converted into heat, the waste heat recovery process is embedded in each IDC to demonstrate the effectiveness of the repurposed heat, which can be used by the residential heating demand in the thermal system, for total cost reduction and energy usage efficiency in the whole operating system. Applying real-life data traces of the electricity price, renewable energies and heating demand, these extensive evaluations demonstrate that both spatial and temporal complementary attempts on the supply side and demand side, along with power and thermal complementary efforts, can significantly reduce the overall cost for the IDC enterprise.