Topological Design of Mesh Communication Networks Using Multiobjective Genetic Optimisation Rajeev Kumar, S. Prasanth, and M.S. Sudarshan Department of Computer Science & Information Systems Birla Institute of Technology & Science, Pilani – 333 031, India rajeevk@bits-pilani.ac.in Abstract. Designing an optimal communication network is a complex, multi- constraint and multi-criterion optimisation problem. Previous work in this field has optimized a single objective or a combination of multiple values into a sin- gle scalar value. In this work, we use the multiobjective genetic optimisation technique along with the convergence to obtain a Pareto front – a set of solu- tions, which are nondominated with respect to a set of constraints and non- inferior to each other – for the network design problem. 1 Introduction Computer and telecommunication network topology design is affected by various factors like network cost, average packet delay, network throughput, reliability of the network and the maximum link capacities. Typically, a network design targets reduc- tion in cost and network delays and increasing the throughput. However, reliability and the maximum possible traffic in a link are the additional requirements to be satis- fied. Thus, network design requires simultaneous optimisation of conflicting factors, subject to various constraints. For example, reducing the packet delay could mean an increase in the link capacities, which will result in an increase in the network cost. Similarly, network reliability means that failure of one node should not disrupt the entire network. A spanning tree network connecting all the nodes, with no additional links, achieves minimal cost but fails to meet the required reliability constraint. Ex- ploring the whole solution space for such a design problem is NP-complete, and heu- ristics have to be relied upon to solve practical problems. Conventional solutions to this problem have been based on rigorous mathematical programming, queuing and network flow concepts [2]. Alternately, heuristic tech- niques like the Branch X-Change (BXC) method, the Concave Branch Elimination (CBE) method and the Cut Saturation method have also been used. However, some of these heuristic solutions have their own limitations. For example, the BXC method is computationally expensive, and the CBE method does not provide for insertion of new links. Previous work in the filed of network design using genetic algorithms, has opti- mised only single, ad hoc objectives and the eventual solutions have been strongly influenced by the linear coefficients used to combine the different (sub-) objectives