On the design and performance of cognitive packets over wired networks and mobile ad hoc networks

This dissertation studied cognitive packet networks (CPN) which build networked learning systems that support adaptive, quality of service-driven routing of packets in wired networks and in wireless, mobile ad hoc networks.; The first part of this study completed the design of the CPN algorithm for wired systems by revisiting the decision making procedure employed by smart packets in the former initiative. To conduct evaluation studies, this work proposed two implementation architectures, both of which were integrated into the Linux 2.4 kernel.; The second part of this dissertation extended CPN to ad hoc networks by introducing mechanisms to handle node mobility with a new quality-of-service metric that makes better use of the limited resources typically present in such networks. In ad hoc CPN (AHCPN), path availability quantifies a path via the probability to find nodes and links available for routing; it is defined as a function of battery lifetime at nodes and signal-to-noise ratio of communication channels. As a result, nodes which have low power are able to extend their working lifetime and routes are selected based on power considerations, leading to the establishment of more robust communication paths over time.; This research included the deployment of three network testbeds and extensive experimentation to verify the performance of both CPN and AHCPN. In addition, a simulation model of AHCPN was developed using Network Simulator 2 (NS-2). Measurements on real systems and simulations revealed great capacity for adaptability and successful performance of both CPN and AHCPN under different network conditions.