Distributed Virtual Network Embedding

The Internet in its current form is considered to be inflexible and difficult to change. Though a number of innovations and improvements (like DiffServ/IntServ, IPv6, ...) have been proposed, the actual architecture of the Internet has experienced a very slow adoption of new ideas. Network virtualization has been recognized as a tool to overcome this perceived ossification [1, 2]. Network virtualization applies virtualization concepts to network resources in order to provide an abstraction from specific hardware. By applying this principle, more flexible virtual resources are created from the underlying physical resources. The added flexibility manifests itself in the possibility to create arbitrarily structured virtual networks that do not necessarily have to reflect the topology or the properties of the underlying physical network. Networks, in this context, are considered to consist of routers and the links that interconnect them. Any number of virtual routers, arbitrarily interconnected, can then be run on a physical topology, provided there are enough physical resources in order to realize the respective virtual resources. The problem of mapping virtual resources to physical resources in such an environment is commonly known as “Virtual Network Embedding” (VNE). A number of algorithms to solve this problem have been proposed in the literature, so far. Most of these algorithms rely on a centralized entity that will compute an optimal or near-optimal solution for a given situation. However, in realistic scenarios a centralized entity should be avoided in order to not create a single point of failure. Moreover, since the VNE problem tends to be computationally expensive, the exploitation of parallelity is expected to improve performance. Thus, it is necessary to investigate novel approaches for a distributed VNE solution.