Routing Connections With Differentiated Reliability Requirements in WDM Mesh Networks

Reliability has been well recognized as an important design objective in the design of modern high-speed networks. While traditional approaches offer either 100% protection in the presence of single link failure or no protection at all, connections in real networks may have multiple reliability requirements. The concept of differentiated reliability (DiR) has been introduced in the literature to provide multiple reliability requirements in protection schemes that provision spare resources. In this paper, we consider the problem of routing connections with differentiated reliability in wavelength-division multiplexing (WDM) mesh networks when backup sharing is not allowed. Our objective is to route connections with minimum network cost (e.g., network resources) while meeting their required reliability. We assume connections arrive dynamically one-at-a-time and a decision as to accept or reject a connection has to be made without a priori knowledge of future arrivals. Since sharing cannot be used for achieving efficiency, the goal is to achieve efficiency by improved path selection. In this paper, we first present an integer linear programming (ILP) formulation for the problem. By solving the ILP formulation, we can obtain an optimal solution with respect to the current network state for each dynamic arrival. To solve the ILP formulation, however, is time consuming for large networks. We thus propose two approximation algorithms for the problem. The first one, called Shortest-Path-Pair-based Auxiliary graph (SPPA), can obtain an epsiv-approximation solution whose cost is at most 1 + epsiv times the optimum in O((n²(n + 1) + 2mn)(log log(2n) + 1/epsiv)) time, where n and m are the number of nodes and links in a network, respectively. To reduce the computational complexity of the first algorithm, the second algorithm, called Auxiliary graph-based Two-Step Approach (ATSA), is proposed and can obtain a near optimal solution with cost at most 2 + epsiv times tha- - t of the optimal solution in O(mn(log log n + 1/epsiv)) time. Results from extensive simulations conducted on two typical carrier mesh networks show the efficiency of the two algorithms.