Multi-Agent Software Defined Network controller: An approach to disaggregate the control plane

Publication Type:
Thesis
Issue Date:
2023
Full metadata record
This thesis proposes a modular Software Defined Network controller - (Multi-agent SDN controller). The thesis is in two parts. The first part makes the proposition and provides the necessary background information. The Software Defined Network paradigm deviates from traditional networks by logically centralising and physically separating the control plane from the data plane. Currently, SDN control planes are monolithic, OpenFlow dependent and do not support application portability. A modular architecture is proposed in the second part of the thesis to improve the existing control plane. The architecture comprises social agents. The architecture varies from reactive to learning agents based on the agent’s role and a knowledge base to formally represent knowledge in a multi-agent system and provide simple inferences. For information exchange, the agents exchange messages. A prototype is designed and built to demonstrate controller operation in some common network scenarios. Performance is evaluated at both the agent and system levels. While it was observed that agent system performance compares well with a monolithic controller, the agent system has higher latency when physically distributed.
Please use this identifier to cite or link to this item: