Résumé en anglais

With the worldwide acceptance of the Internet as a de-facto public network, there is great interest in the deployment of Virtual Private Networks (VPNs) across IP networks. VPNs provide an economic alternative to the expensive private leased lines. The main purpose of a VPN is to provide a company secure communication among multiple sites through the shared Internet. The allocation of bandwidth for VPNs to meet the requirements specified by customers is an important traffic engineering research issue. This paper addresses the problem of provisioning VPN services based on pipe and hose workload models over shared network infrastructure with bandwidth guarantees, an issue studied only by different resource allocation methods [1] and not studied before by general optimization methods. The general problem of computing a constrained VPN tree with optimum bandwidth allocation is a hard combinatorial optimization problem [2]. Our proposed optimization method for provisioning VPNs is based on the Simulated Evolution (SE) meta-heuristic originally introduced in [3]. The proposed Evolutionary Algorithm (EA) connects VPN nodes using a tree structure while seeking to optimize the total bandwidth reserved on the edges of the VPN tree. Experimental results with Waxman network graphs [4] show that tree bandwidth costs obtained with hose workloads are higher by a factor of a maximum 2.5 compared to those obtained with pipe workloads.