Optimizing the embedding of virtualized cloud network infrastructures across multiple domains

Network is now a main part of the cloud resources, becoming more than a connectivity add-on. The ability to define network resources (e.g. routing/switching elements, bandwidth, delay) in this context is still scarce, but there are clear evidences that this is a future reality, as is the case of Network Functions Virtualization (NFV). In this paper we consider that cloud infrastructure services will allow the definition of complete infrastructures, to which we refer as Virtual Infrastructures (VIs). These VIs comprise both computing, storage and network resources. This paper specifically tackles the VI embedding problem in scenarios with multiple domains. These scenarios consider multiple Data Center (DC) locations inter-connected through an operator network. The embedding problem is known to be NP-hard, and therefore we present an embedding strategy based in Integer Linear Programming (ILP) formulation. The proposed strategy aims to balance the load among the different domains, and considers the location as a key constraint for virtual resources. Finally, a thorough evaluation of the formulation is performed, analyzing the VI acceptance ratio and the occupation of the physical infrastructure. The obtained results show that the location constraint has a high impact on both the acceptance ratio and occupation of physical resources.