Simulated evolution based algorithm versus exact method for virtual private network design

Abstract

In this paper, we studied Virtual Private Network Design Problem using tree structure and assuming a pipe traffic matrix. Today with network virtualization, Virtual Private Networks (VPNs) have more importance in networking and offers the company the ideal solution to establish ondemand overlay networks that enable their customers to securely access company resources. This is a hard combinatorial optimization problem that it has been solved in literature only with approximation methods. Generally, this kind of methods do not give any guarantee on the solution quality and we are enable to know a priori how far the given solution is from the optimal one unlike other methods such as the exact methods. For this purpose, we propose an integer linear program (ILP) with classical Pipe traffic model to design the problem under investigation. Based on the proposed integer programming formulation, we solve the problem using two approaches: The first contribution is Simulated Evolution based evolutionary algorithm and the second contribution is an exact method based on Branch and Cut (B&C) algorithm to find a tree rooted at a user specified node with minimized overall reserved bandwidth. Performance results using Brite networks show that our proposed evolutionary algorithm offers good solutions within a fraction of the time required by the B&C algorithm and bandwidth cost within at most 1.5% of the optimal solutions found by the exact method.