On DiffServ traffic forecast and queue management using neural network with logarithmic scaling and generalized quantization

In this paper we propose a new predictor based on feed forward neural network featured by logarithmic scaling and generalized quantization procedure. We analytically prove its optimal behavior. Further, we integrate our predictor into a queuing system framework with control that serves DiffServ traffic. Within that framework we introduce the SCPO logic which is an essential part of the queuing system performance optimization task. Using simulation studies we show efficiency of the proposed SCPO logic as well as of the designed predictor.