Online Virtual Network Embedding Based on Virtual Links’ Rate Requirements

Virtual network embedding (VNE) is the process of allocating substrate network resources to support virtualized networks (VNs), mapping each virtual node to one substrate node and each virtual link (VL) to one or more substrate links. VLs are mapped onto one or more substrate links in which, for the latter case, substrate links are connected through intermediate substrate nodes (ISNs). Current VNE approaches ignore the CPU capacity that ISNs require to forward traffic on the VLs mapped onto them while computing the VN embedding cost. We demonstrate that the CPU processing of ISNs required to forward VLs' traffic is not negligible or constant and that, it is rather a function of the rate requirements of the VLs mapped onto them. We formulate the VLs' forwarding rate requirements function (VLR2) to determine the CPU utilization that ISNs would require to forward the traffic at the rates required by the VLs. Based on the VLR2 function we formulate the VNE-VLR2 approach, which extends the VNE problem to a more realistic approach, where the VN embedding cost also considers the CPU needed to forward the VLs' traffic at the required rates. Rather than proposing performance enhancements to current VNE approaches, this paper demonstrates that there are sensible differences in the performance of the VNE problem while considering the ISNs' CPU utilization due to the VLs' forwarding rate requirements. The differences of our VNE-VLR2 approach with respect to current VNE approaches in key metrics like VNE acceptance ratio, revenue-to-cost ratio and convergence time highlight the importance of this practical consideration for more accurate, realistic and quality of service-aware online VNE processes.