Differentiating complex network models: An engineering perspective

Network models that can capture the underlying network's topologies and functionalities are crucial for the development of complex network algorithms and protocols. In the engineering community, the performances of network algorithms and protocols are usually evaluated by running them on a network model. In most if not all reported work, the criteria used to determine such a network model rely on how close it matches the network data in terms of some basic topological characteristics. However, the intrinsic relations between a network topology and its functionalities are still unclear. A question arises naturally: For a network model which can reproduce some topological characteristics of the underlying network, is it reasonable and valid to use this model to be a test-bed for evaluating the network's performances? To answer this question, we take a close look at several typical complex network models of the AS-level Internet as examples of study. We find that although a model can represent the Internet in terms of topological metrics, it cannot be used to evaluate the Internet performances. Our findings reveal that the approaches using topological metrics to discriminate network models, which have been widely used in the engineering community, may lead to confusing or even incorrect conclusions.