Provisioning virtual private networks in the hose model
Haibo Wang · 2007
Virtual private networks (VPNs) can provide customers with cost-effective, reliable, and secure services.Currently there are two resource management models for provisioning VPN services, the pipe model and the hose model.The hose model only requires customers to specify an aggregate ingress and egress bandwidth capacities per-endpoint, instead of an exact traffic demand per endpoint pair as with the pipe model.This thesis investigates hose model VPN provisioning algorithms on the following important but neglected aspects.Firstly the blocking performance of the hose model and the pipe model in provisioning bandwidth-guaranteed connections are analyzed and compared, using both analytical models and discrete-event simulations.This study shows that the hose model has a blocking performance better than the pipe model by several orders of magnitude under normal network operating loads, contributing to the justification of using the hose model as the preferred resource management model for VPN service provisioning.Secondly the hose model can be implemented in tree routing, single-path routing and multi-path routing.In this part, a thorough study of different routing schemes for the hose model is presented, from the cost as well as the blocking performance perspectives.In addition, a new concept of sub-provisioning is proposed to overcome the bandwidth overprovisioning problem associated with the hose model.The results show that compared with tree routing, multi-path routing only brings in marginal overprovisioning at rare combinations of network topologies and hose parameters, but it delivers a much better blocking performance at a fixed amount of provisioning.