Evaluation of Performance of Genetic Algorithms for Network Tomography

Wide Area Network monitoring has become increasingly important to deliver Quality of Service (QoS). Fortunately recent research has devised interesting approaches for scalable network monitoring using Linear Modeling. However, such linear models involve an underdetermined or over-determined system of equations with no unique solution. Finding an optimum solution in such scenario involves posing the problem as a constrained optimization problem. However, such constructions can involve a variety of options of objective function selection. For example one approach in objective function selection is driven by pre-specifying traits in the desired solution; other approaches pose no such constraints. More recent approaches combine the merits of various approaches for best optimum solution. In this paper we propose the use of genetic algorithms to solve these optimization problems since they provide an ideal platform in using multiple pronged objective functions to bridge the dichotomy between the various methods available and come up with the best solution. Our findings in this paper are that genetic algorithms can surpass the conventional approaches (of convex optimization) in solution construction and often surpasses performance of conventional approaches while having acceptable computation times.