Optimization of network clustering and hierarchy through simulated annealing

Abstract

In recent years, there are several design issues in computer networks such as clustering and hierarchy, which have been in the spotlight. This is due to the desires to improve the network performance and scalability. The clustering problem is to partition the network's clients into a minimal set of clusters to maximize the local cluster traffic. The hierarchy problem is to connect all clients' clusters into a minimal network levels so that the network scalability and security are maximized. We have combined the two problems (clustering and hierarchy) into a single optimization problem, where we are seeking the minimal cost of network topology with the best possible clustering and hierarchy. The network cost depends heavily on the selection and placement of the network devices within the clusters. Because of the combinational nature of these problems, this combined optimization problem is suitable to be tackled with a meta-heuristic stochastic search method, such as Simulated Annealing (SA). Our experimental results have revealed that the efficiency of our proposed search method in finding good solutions within few minutes.