Optimizing linear prediction of network traffic using modeling based on fractional stable noise

Reliable prediction of network traffic allows for the implementation of more efficient resource management schemes. In a previous work, reported by some of the same authors of this paper, a novel algorithm for linear prediction of network traffic was introduced and evaluated. That algorithm assumed that traffic statistics can be modeled using alpha-stable long-range-dependent stochastic processes. The relevant prediction algorithm was based on the minimum dispersion criterion, whose resulting equations were solved in a processing-efficient but approximate manner. More recent work has proved that in most of the cases the coefficients so obtained produce a robust and acceptable performance. Nevertheless, further studies suggest that the accuracy of the linear prediction can be enhanced if needed. This work identifies where this can be done, proposes some optimization procedures and provides some numerical examples. Our results show that, when incorporating optimization, the gain in performance is quite remarkable for network traffic exhibiting strong long-range dependence