An artificial neural network approach for loss estimation in bufferless optical packet switched networks

Abstract

Optical Packet Loss Rate (OPLR) is a critical performance measure for Optical Packet Switched (OPS) networks. In this paper, we use a common Artificial Neural Network (ANN), that is multilayer perceptron (MLP), as a fast OPLR estimator for bufferless slotted OPS networks. Based on the proposed estimator, an algorithm called OPLRC is proposed to control Optical Flow's Packet Loss Rates (OFPLRs) which are OPLRs for optical packets transmitted from different ingress switches. OPLRC is a greedy algorithm which uses estimated OFPLRs of the trained MLP to tag some optical packets at ingress switches as eligible for drop at core switches in case of contention. This will control OFPLRs of un-tagged optical packets at the specified limits while giving some chance for tagged optical packets to reach their destinations. The accuracy of the proposed estimator is analyzed using numerical methods and extensive simulations. Results show that OPLRC can manage OFPLRs of untagged optical packets with an acceptable accuracy.