Dynamic virtual network dimensioning in cost-sensitive environments

Virtual private networks (VPN) over ATM or IP-based public networks provide significant cost advantages compared to fixed-bandwidth private line services. This work studies the VPN dimensioning problem in an environment where the public network prices connection-oriented services based on source, destination and grade of service, and advertises these prices to its VPN customers (users). Users then decide how much capacity to allocate based on a local utility function and routing constraints. Traffic between local networks can follow a direct path, or be routed through intermediate nodes of the VPN, or simply use the public network's transport service. Using this flexible routing scheme, the VPN operator can adjust the size of VPN links to reach a minimum cost topology. The latter is produced by running a VPN capacity optimization algorithm. Our findings are extended to an environment where many service classes are supported on distinct virtual links (and hence requiring one VPN per service class), or on a single virtual link whose capacity is characterized by a multidimensional capacity region. We apply the algorithm to a sample network and illustrate the results.