Towards provisioning hybrid virtual networks in federated cloud data centers

Abstract

Network virtualization is an efficient way to enhance the resource utilization of physical network. It enables numerous heterogeneous virtual networks (VNs) coexist and share the resources of same physical network. Virtual network provisioning has been a key issue in network virtualization. Since the optimal virtual network provisioning is an NP-hard problem, existing studies devote to propose heuristic approaches for a tradeoff between computational complexity and the quality of VN provision. A traditional physical/substrate network usually sustains numerous infrastructure providers (InPs), and many applications in the substrate network can be characterized by hybrid virtual network which composed by both unicast and multicast virtual network. However, few research has conducted for the problem of hybrid virtual network provisioning (HVNP) among multiple domains. In our research, we model the HVNP problem through integer linear programming (ILP) for minimizing provisioning cost. Furthermore, we propose two effective algorithms to address the researched problem: (i) the decomposition-based algorithm, HVNP_D; and (ii) the spectral clustering based algorithm, HVNP_SC. Extensive simulation experiments have been carried out to assess the proposed algorithms. Simulation results demonstrate that our approaches have better performance than existing approach.