Standalone Data-Center Sizing Combating the Over-Provisioning of the IT and Electrical Parts

Abstract

During the two last decades, sustainability of IT infrastructures like datacenters became a major concern for computer giants like Google or Amazon. Datacenters powered with renewable energy have been proposed. But because of the intermittency of these power alternatives, these platforms remains connected to the classical power grid. IT structure and electrical constraints were often questioned separately, leading to a non-efficient global system. In this paper, an energy self-sufficient green datacenter is modeled and designed, by proposing an electrically autonomous infrastructure including wind turbines, solar panels and its own short and long-term storage devices mainly based on batteries and hydrogen system. Existing sizing approaches limit themselves to a perfect QoS leading to an overestimation of the needed equipment. In this paper we show how to reduce and combat this overprovioning by questioning its impact on the QoS and needed equipment: decreasing the number of computing or storage elements (servers and batteries). As an example, decreasing the targeted QoS from 100 to 95% more than halves the needed number of servers, while a decrease of 30% of the battery capacity has a negligible impact on the electrical infrastructure.