ºìÐÓÊÓƵ

Topic

FTRL-WRR: Learning-Based Two-Path Scheduler for LEO Networks

Department of Computer Science

Date & location

• Friday, November 22, 2024
• 10:00 A.M.
• Engineering Computer Science Building
• Room 467

Reviewers

Supervisory Committee

• Dr. Jianping Pan, Department of Computer Science, University of Victoria (Supervisor)

• Dr. Jaya Prakash Champati, Department of Computer Science, UVic (Member) 

External Examiner

• Dr. Hong-Chuan Yan, Department of Electrical and Computer Engineering, University of Victoria 

Chair of Oral Examination

• Dr. Dante Canil, School of Earth and Ocean Sciences, UVic

   

Abstract

Multipath QUIC is inspired by the resource pooling principle, aiming to make a collection of resources behave as a single pool. However, current multipath schedulers tend to prioritize specific metrics like RTT or congestion window, often overlooking strategies that enhance overall resource usage and reduce completion time. This can lead to resource underutilization in high dynamic settings, such as those involving LEO satellites. Addressing this challenge requires efficient traffic allocation to maximize bandwidth utilization. In this thesis, we verify that the relationship between traffic distribution and throughput in a two-path scenario resembles a quasi-concave function. Accordingly, we formulate the traffic allocation across two paths as a 1-dimensional optimization problem. To solve the two-path scheduling problem in dynamic environments, we introduce the FTRL-WRR algorithm. This approach integrates an FTRL learning module, ADWIN2 tuning, and WRR scheduling to enhance bandwidth utilization. We validate the effectiveness of the algorithm through extensive emulation and real-world testbed experiments, demonstrating consistent re duction in completion time across a range of scenarios. Additionally, we discuss the algorithm’s limitations and suggest directions for future research.