Interconnecting LANs and a FDDI backbone using transparent bridges: A model and solution algorithms

In this article, we construct a mathematical model to design the topology of a network interconnecting a number of local area networks (LANs) and a fiber-distributed data interface (FDDI) backbone using IEEE 802.1 transparent bridges. The model allows parallel bridge links between subnets and yields a two-node-connected topology for survivability. The model determines the routing tree on a given topology according to the IEEE Standard protocol, which uses as input only user-supplied parameters such as bridge IDs, port IDs, and path costs. The model ensures that none of the network components (LANs, FDDI, and bridge links) are congested even in the failure of a LAN or a bridge link. We develop a path cost adjustment algorithm and a tabu search algorithm to find a low-cost feasible topology. Through computational experiments, we identify the characteristics of feasible topologies found under varying traffic conditions. Under light traffic conditions (i.e., when LANs' total slack capacity is large relative to the total intersubnet traffic demand), the topology is close to a ring which costs much less than the conventional star topology. Under heavy traffic, the topology tends to be a ring augmented by parallel bridge links between LANs and the FDDI. Finally, we discuss how effective various heuristic ideas – such as uncapacitated, two-connected topology design heuristics, the short-term tabu list and the long-term tabu list – are with respect to improving the solution quality. We also present a Lagrangian relaxation-based lower bounding procedure and report the gap between the best feasible value and the lower bound. The parsimonious property of uncapacitated, two-connected topology design problems is employed in this dual partitioning procedure.