Effects of variations of load distribution on network performance

The paper is concerned with the characterization of the relationship between topology and traffic dynamics. We use a model of network generation that allows the transition from random to scale free networks. Specifically, we consider three different topological types of network: random; scale-free with /spl gamma/=3; scale-free with /spl gamma/=2. By using a novel LRD traffic generator, we observe best performance, in terms of transmission rates and delivered packets, in the case of random networks. We show that, even if scale-free networks are characterized by shorter characteristic-pathlength (the lower the exponent, the lower the pathlength), they show worst performances in terms of communication. We conjecture that this can be explained in terms of changes in the load distribution, defined as the number of shortest paths going through a given vertex. In fact, that distribution is characterized by (i) a decreasing mean and (ii) an increasing standard deviation, as the networks becomes scale-free (especially scale-free networks with low exponents). The use of a degree-independent server also discriminates against a scale-free structure. As a result, since the model is uncontrolled, most packets go through the same vertices, favoring the onset of congestion.