Keywords

Software-Defined Networking, Reinforcement Learning, Traffic Flows, Low-latency Applications

Abstract

Software-Defined Networking (SDN) separates the control and data planes, enabling better programmability of the control plane to predict, route, and schedule traffic at the data plane. Reactive SDN dynamically installs flow rules when a new flow arrives, making it adaptable to application dynamics. This design is well-suited for emerging low-latency applications like online gaming and AR/VR, which require millisecond-level response times for an acceptable quality of experience. However, Reactive SDN faces limitations, as each new flow triggers a miss, sending a Packet-in message from the switch to the SDN controller, increasing latency. To reduce delay, it's crucial to predict flow arrivals and install the necessary rules in advance. Reinforcement Learning (RL), where agents make decisions based on rewards, can help by learning to predict flow arrivals and installing flow rules ahead of time. We propose a Speculative SDN design to address the limitations of Reactive SDN for low-latency applications. This approach uses RL to predict and install unseen flows, reducing delays and improving overall performance. By speculatively installing flow rules, we enhance SDN's ability to meet the fast-response demands of emerging applications.

Completion Date

2025

Semester

Spring

Committee Chair

Yuksel, Murat

Degree

Doctor of Philosophy (Ph.D.)

College

College of Engineering and Computer Science

Department

Computer Science

Identifier

DP0029315

Document Type

Dissertation/Thesis

Campus Location

Orlando (Main) Campus

Download

Share

COinS