Reduced complexity algorithms for cognitive packet network routers

Abstract

The cognitive packet network (CPN) routing protocol provides a framework for real-time quality of service decision making within packet networks. This allows the paths taken by packets to autonomously adapt to changing conditions in order to maintain and improve on the quality of service provided by current routing algorithms. Software implementations of the protocol use the random neural network with reinforcement learning. This algorithm is unsuitable for implementation in dedicated hardware or devices with low computational abilities due to its complexity. We present a series of alternative algorithms for use in CPN, and compare their complexity and performance with respect to software and hardware implementation. Through experimentation we demonstrate that it is possible to match the performance of the random neural network with the simpler alternative algorithms. We also propose an architecture for an FPGA based hardware CPN router, and describe our implementation.