As computer communication networks grow rapidly, a key challenge is how to manage large-scale heterogeneous networks for supporting diverse applications. I will first provide an overview on fundamental issues, and our approaches on network management. I will then focus on how to tackle the challenges due to the heterogeneous traffic by presenting our recent results.
In particular, the challenge originates from heterogeneous network traffic at different granularity as a complex mixture of long-range and short-range correlations. Such network traffic invalidates previously developed traffic models, and makes network management and control a difficult task. A significant discovery from this work is that the network traffic is no-longer long-range correlated in the wavelet domain. Therefore, independent or simple Markov models can be used to characterize network traffic in the wavelet domain. This opens up new possibilities for modeling, performance analysis, and network control to be done in the wavelet domain rather than in the time domain.