Abstract
Since its commercialisation in 1995, the Internet has grown exponentially. Cisco estimates that the number of Internet users will increase to 5.3 billion in 2023 from 3.9 billion as of 2018. To confront the issue of increased Internet usage, new, smarter, and more versatile networking solutions are required.Software Defined Networking (SDN) is a relatively new paradigm that introduces a centralised approach where the intelligence of each network device is stored in a central SDN controller. SDN offers several advantages over distributed networking including, faster adaptation to network topology changes, easier integration of network applications, simultaneous control of many network devices, etc. However, despite these advantages, SDN traditionally uses heuristic algorithms such as single metric shortest path which may not be suitable for Quality of Service (QoS) based routing.
The aim of the research was to create, implement and test a routing algorithm as an alternative to existing heuristic algorithms. Q-Routing was selected as the basis of a new routing algorithm as literature shows that Q-Learning holds several advantages over other reinforcement learning algorithms while being faster at pathfinding than shortest path in distributed networks. The Q-Routing based algorithm had to meet several criteria: i) Use the advantages presented by SDN. ii) Employ multiple network metrics for QoS based routing. iii) Perform as well as K-Shortest Path. iv) Find paths faster than K-Shortest Path. v) Adapt to dynamic network environments. vi) Work efficiently in network topologies of different sizes and densities.
Research was split into three phases, initially looking at SDN based single metric Q-Routing leading to the final algorithm, Dual Metric, Multi-Estimate, Pareto, Q-Routing which successfully met all stated goals. Emulated results show that the algorithm in terms of the percentage of flows blocked performed as well as K-Shortest Path while being faster at pathfinding in both static and dynamic network environments.
| Date of Award | 25 Jan 2022 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Trevor P Martin (Supervisor) & Reza Nejabati (Supervisor) |