Utilizing neural networks to reduce packet loss in self-similar teletraffic patterns

Reducing packet loss and increasing overall efficiency in multiple source queuing systems is one of the most important issues in the design of traffic control algorithms. On the other hand, the other important issue in such systems is to provide every individual source with the ability to take advantage of a fair portion of the shared available resources such as buffer space or server bandwidth. In this paper a novel technique for reducing packet loss in a class of queuing systems with self-similar traffic patterns is introduced. The technique takes advantage of the modeling power of neural networks to offer a dynamic buffer management scheme capable of efficiently addressing the trade off between packet loss and fairness issues.