Today’s data centers take an ever growing share of the energy consumption of Germany's ICT. The main goal of this work is to reduce such part of the consumption by minimizing the number of active servers required to deploy a given set of services.
This approach belongs to the field of resource management. In contrast to known work, the new approach takes the resource demand behavior of the services into account in special way. Hence, an important part of this work is to properly model this behavior. In addition, it is supported to trade off resources against service performance. Finally, the algorithms that assign the services to servers form another challenge that is addressed within this thesis.
The developed approach was evaluated based on simulations. Already the initial distribution of services to servers achieved between 12.5% and 42% of energy savings. The consumption could be further reduced by additional 20%, when services are dynamically redistributed at runtime.