| In this paper, we consider bandwidth-intensive services for customers that want virtual networks (VN) in a data center environment. In particular, we consider this problem in a temporal context where bandwidth-intensive requests from each VN may arrive randomly at a review point, which may last for a certain duration. Thus, at each review point, the data center network provider must optimally allocate resources for the demand requests. For this problem, we present a mixed-integer programming (MIP) problem formulation where any request from a VN customer may be assigned to any virtual machine so that network resource availability is optimized. We present an overbooking strategy that may be employed to allow for some demands not met in the first try. For comparison, we also consider a base case where the allocation is pinned to a specific destination. Through our study, we show the comparative gains of different schemes. |
