Service Deployment in Heterogeneous Cloud-like Environments

The amount of power consumed by data centres world wide is increasing, and due to growing electricity bills, service providers aim more attention on energy-efficient management of their data centres. In order to achieve this goal, a service provider need to make smart decisions regarding the deployment of his services. At the same time, in order to satisfy his end-users, a service provider needs to focus on delivery of services complying with the quality of service (QoS) requirements. Consequently, he needs to make decisions related to replication level of his services, as well.In this thesis, I propose two interrelated mixed integer linear programming (MILP) models aiming at supporting service providers in their decision making. The first MILP concerns energy-efficient deployment of a service provider's services in his own virtualized data center, where the objective is to minimized the cost of energy usage, while satisfying the response time and availability requirements of the end-users. The second MILP introduces the flexibility of Cloud computing by letting the service provider have the opportunity to deploy services in a public cloud, and hence the objective is to minimize the total cost of deployment, while still, ensuring satisfactory QoS levels. The proposed MILP models are tested on test instances of varying size with the intention to discuss scalability issues and commenting on modelling choices. The results show that the second model is the hardest to solve, in terms of closing the optimality gap, but nevertheless, it is depicted that good solutions are found early in the branch and bound search. Furthermore, different modelling choices illustrate the trade-off between energy-efficient management of data center resources and service performance