Proceedings of the 28th EnviroInfo 2014 Conference, Oldenburg, Germany
Rising power consumption of data centers is a topic of great concern and therefore several power
saving technologies exist. This paper describes the idea of a data center overall power saving and
controlling strategy, allowing the data center to enter optimized minimum power states but also to
control its own power consumption to apply demand response management. Therefore, the degrees
of freedom a virtualized data center has are modelled and the methodology used to control its
energy state is described, taking into account the IT hardware like servers and network gear as well
as the influence of cooling devices and power distribution devices. In the paper, we describe our
models used for our simulations, the methodology and the power saving potential of our system.
We formulate the problem to control the data center’s power consumption by applying different
consolidation strategies as an extended bin packing optimization problem, where virtual machines
must be packed on a specific number of servers. External constraints like the time-flexibility of the
solution and the influence on supporting devices are applied by using cost functions. We present a
greedy solver for this problem and show first results and the potential of the entire approach.
9 / 2014
AC4DC - Adaptive Computing for Green Data Centers Steigerung der Energieeffizienz durch intelligentes Rechenlast- und Infrastrukturmanagement vom Anbieter bis zum Anwender