Everyday economic and social wellbeing depends on reliable, secure and efficient data centers and the services they provide. At the same time, just in the EU data centers consume approximately 11.8GW (or 103,368 GWh p.a.) which is around 3\% of the total electricity generated across the EU. The datacenters industry market of €18.85 billions p.a., with a generation of 38.6 million tonnes of CO2 p.a. (comparable to the aviation industry). The societal need is thus to make data centers the most energy-efficient possible.
Scope of the work
Among the various ways to increase data centers' efficiency, here we focus on how to control IT loads and cooling loads within a data center so that to avoid energetic waste. More specifically, we focus on jointly solving the problems of where / when to turn on and off servers and virtual machines, where / when to allocate IT loads, where / when / how much to cool the servers by means of the air conditioning system.
Description of the work
The work wants to improve upon the state of the art by using stochastic model predictive approaches to solve the problems above. In practice, we aim at using what is considered the novel state of the art control technique, something that up to now has been investigated by researchers only from theoretical perspectives and rarely applied to real systems.
The student will thus learn the fundamentals of this novel theory, and apply it to the data centers case. Applications will be both on simulators and real systems: experiments will be performed in collaboration with SICS (the Swedish Institute of Computer Science) on the SICS-ICE data center.
The student will be supervised by two faculties and one PhD student.
In case of successful results, we aim at publishing the results in some major international journal.