Routing and Provisioning VPNs based on Hose Traffic Models and/or Constraints

In this paper we investigate various questions associated with routing and provisioning virtual private networks (VPNs) based on hose traffic models or constraints, i.e., using aggregate (hose) traffic characterizations in/out of VPN end-points. This appears to be a promising simple approach to dealing with traffic engineering for VPNs. We evaluate the possible efficiencies derived from spatial multiplexing, i.e., over possibly varying routings of traffic loads consistent with hose specifications, and temporal multiplexing, i.e., statistical multiplexing of bursty loads, on shared VPN resources. We explore how these impact VPN routing and provisioning to meet ‘coarse grained QoS targets’ and how routing a sequence of VPNs on a shared network might be done to extend the ‘life-time’ of these decisions (i.e., so as to avoid re-routing) as the number of, and demands on, VPNs grow. Finally we discuss some natural extensions of the hose service model, specifically, the inclusion of multiple classes and cutset constraints, and how application layer multicasting mechanisms might balance loads over a VPN’s hoses.