Abstract-Network virtualization has been proposed as a way to increase the flexibility of the Internet. This could enable the provisioning of many desired services in the current architecture and allow effective sharing and use of resources. Providing virtual networks (VNs) means that virtual nodes and links need to be embedded in the underlying shared infrastructure. This embedding process, where VNs with resource demands are mapped onto a substrate with finite resources is a challenging and NP-hard problem. In this paper the focus is on mapping the VNs in such a way that node resources in the substrate are not completely exhausted. To achieve this objective, an approach referred to as bottleneck node reduced mapping is presented. This method is evaluated and compared with an approach, where resource exhaustion is not considered. [7,9,11]. It can be utilized in experimental research facilities [13, 14,15] as well as in provision of customized end-to-end services over a shared infrastructure [7,9]. A number of virtual networks (VNs) can be deployed on top of the physical network (or substrate), depending on the capability of the substrate and the demands of VNs.
KeywordsThe virtual network embedding (VNE) problem is NPhard [8,17] where several constraints need to be satisfied. In order to map a VN onto the substrate, requirements of both its vertices as well as edges should be fulfilled. In addition to this, VNs can arrive at different times, in any order and can be based on any standard network topology (e.g. star, bus, ring or mesh). The substrate network also has a limited amount of resources. Thus, we need to embed or map a VN with resource constraints onto the substrate network (SN) which has finite resources.In the past, various efforts have been made to suggest solutions for embedding VNs onto physical networks. By considering that substrate resources are finite, we make "admission control" an integral part of our solution in contrast to previous solutions proposed in [4,6,12,18]. Our solution considers that VN requests are dynamic and are received online, which is contrary to several other solutions [6,12,18]. Some previous work points to specific virtual topologies [12], whereas our solution is not limited to any specific topology. Our approach is inspired to some extent by [8] as we use similar notations to denote both virtual and substrate networks. There are solutions with support for path splitting [10] as well as those considering path migration in the physical network
