Virtual Network Embedding: Reducing the Search Space by Model Transformation Techniques

Virtualization is a promising technology to enhance the scalability and utilization of data centers for managing, developing, and operating network functions. Furthermore, it allows to flexibly place and execute virtual networks and machines on physical hardware. The problem of mapping a virtual network to physical resources, however, is known to be \(\mathcal {NP}\)-hard and is often tackled by optimization techniques, e.g., by (ILP). On the one hand, highly tailored approaches based on heuristics significantly reduce the search space of the problem for specific environments and constraints, which, however, are difficult to transfer to other scenarios. On the other hand, ILP-based solutions are highly customizable and correct by construction with a huge search space. To mitigate search space problems while still guaranteeing correctness, we propose a combination of model transformation and ILP techniques. This combination is highly customizable and extensible in order to support multiple network domains, environments, and constraints allowing for rapid prototyping in different settings of virtualization tasks. Our experimental evaluation, finally, confirms that model transformation reduces the size of the optimization problem significantly and consequently the required runtime while still retaining the quality of mappings.