Application of Machine Learning Techniques to Delay Tolerant Network Routing
Rachel M. Dudukovich · OhioLink ETD Center (Ohio Library and Information Network) · 2019
DUDUKOVICHThis dissertation discusses several machine learning techniques to improve routing in delay tolerant networks (DTNs).These are networks in which there may be long one-way trip times, asymmetric links, high error rates, and deterministic as well as non-deterministic loss of contact between network nodes, This work focuses on the aspects of scheduling and routing in DTNs.The term scheduling can be taken two ways within this context, on one hand meaning the scheduling of communication between assets such as science mission communication systems, relay satellites and ground stations or the scheduling of messages (packets, bundles, etc.) for transmission and deletion within a single system.Routing focuses on the selection of nodes to form a path from a message source to the destination in the network.These three considerations ( scheduling of communication assets, scheduling of message queues within the communication system, and route selection) are all interrelated and impact one another.The scheduling of assets imposes constraints on what nodes are available for communication at a given time.The scheduling and processing of messages (queuing, re-queuing, expiration, deletion, transmission) impact the traffic in the network, the waiting time for new messages to be sent and received, and the buffer availability on communication assets.Routing impacts the utilization of specific nodes, may cause duplicate messages to exist within the network (based on protocol used) and will influence the overall end-to-end delay of message delivery based on paths selected.This work focused primarily on scheduling of message transmission and routing.1.2.1 Goals of Cognitive Networking for the NASA SCaN Network This section briefly discusses the high level goals for cognitive networking, as outlined by Ivancic, et al. [3] and how they serve as the inspiration for the research focus of this dissertation.• Reducing operations cost: Operations costs are related to the amount of human labor required to support systems operations.The more autonomously the network performs as a whole will reduce labor costs.Automated scheduling of network assets, automated detection of failed nodes and replacement by rerouting or redundancy, and nodes which in" for failed nodes or making predictions and taking proactive measures for when a system failure may occur.• Increase system asset utilization: Machine learning algorithms can be used to augment or as an alternative to human-created or rule based schedules.This may allow for quicker reactions to unanticipated schedule changes or to produce more efficient schedules.DTN testing with 10's of mobile network nodes.Several frequently used simulation and emulation tools including OMNeT++, ONE simulator and CORE/EMANE emulation.• Develop an architecture, problem formulation and considerations for machine learning techniques relative to the environmental challenges