Mapping the virtual networks with stochastic bandwidth resource request in multi-datacenters

Network virtualization allows the design of multiple diverse virtual overlay network architectures over a common physical infrastructure. A key issue in designing such virtual networks (VN) is the mapping of the VN onto the underlying substrate network. Accordingly, many research works have focused on the VN mapping problem. However, these works only consider the case where the VNs require deterministic amount of network resources. Some other works even deal with the dynamic resource demand by using over provisioning, which is simple but inefficient. In this paper, we investigate the online stochastic VN mapping (StoVNM) problem in multi-datacenters, in which the VN requests follow a Poisson distribution and the associate VN bandwidth demands follow a Normal distribution. We formulate the StoVNM problem as an optimization problem with the objective of minimizing mapping cost and load-balancing. Since the VN mapping problem is NP-hard we devise a sliding window technique based on heuristic algorithm w-StoVNM for tackling this NP-hard problem efficiently. Through extensive simulations we demonstrate the effectiveness of the proposed approach compared to traditional VN mapping solutions in terms of VN mapping cost, blocking ratio and total net revenue in the long term.