- This event has passed.
Ph.D. Defense: Timothy Curry
July 31, 2023 @ 10:00 am - 11:00 am EDT
Title: Optimal Routing Configurations for Software-Defined Networks
Ph.D. Candidate: Timothy Curry
Major Advisors: Dr. Benjamin Fuller, Dr. Laurent Michel
Associate Advisors: Dr. Minmei Wang, Dr. Amir Herzberg
Date/Time: Monday, July 31st, 2023, 10:00 am
Location: HBL1102 and WebEx
Password: k6gCRFpax62
Abstract
Network configuration is a crucial, yet delicate, task for any enterprise. When developing a configuration, one must consider multiple aspects such as functionality, performance, and security. Such properties are often conflicting, making it difficult for human network engineers to quickly design a desirable configuration. Current techniques often provide high-level recommendations without supplying the low-level instructions, leaving engineers to manual search for a suitable configuration or use a secondary tool to heuristically find a detailed implementation that may not be fully composable with the original recommendation engine. This paradigm is not ideal nor sustainable for modern networks. This procedure is a perfect application for discrete optimization technologies. One can create a model that ensures certain crucial networking properties are upheld while the other properties are optimized to the specifications needed for any particular use case.
Software-defined networks (SDNs) provide a programmable network infrastructure where global network information can be used to inform routing decisions. Due to their flexibility and programmability, SDNs are well-suited to pair with optimization frameworks that develop routing configurations. This dissertation explores discrete optimization models designed to develop SDN routing configurations. These models automate routing decisions for network flows while mitigating risk due to unwanted or extraneous flows. Network security is considered either jointly with routing or separately in a second stage with a dedicated model. The notion of security is explored over several different metrics, including attack graphs and a novel elastic string trust model. By contemplating both functionality and security, either together or in tandem, the frameworks are able to generate routing configurations to initialize networks from scratch or quickly altering routing in response to suspicious network activity. These approaches are tested on the fat-tree topology, a commonly deployed data center network topology. This thesis explores models of increasing fidelity and realism to demonstrate their applicability in medium-sized networks.
