Optimizing Interconnection Topology using Deep Reinforcement Learning

Graduate Thesis Or Dissertation

Public Deposited

Citeable URL: https://ir.library.oregonstate.edu/concern/graduate_thesis_or_dissertations/mp48sm86s

Descriptions

Attribute Name	Values
Creator	Zeng, Kai
Abstract	As the number of nodes in high-performance computing (HPC) systems continues to grow, it becomes increasingly important to design scalable interconnection network topologies. Prior work has shown promise in adding random shortcuts on top of an existing topology to reduce average hop count and network diameter, but has been limited to naïve and heuristic ways to add shortcuts. In this work, we propose a novel and systematic approach that combines deep reinforcement learning and Monte Carlo tree search. The proposed framework is able to explore the large design space of adding shortcuts effectively. Compared with state-of-the-art topologies for HPC systems, the solution found by our framework significantly reduces the network diameter by 15% and shortens the average hop count by 16%.
License	All rights reserved
Resource Type	Masters Thesis
Date Issued	2021-12-10
Degree Level	Master's
Degree Name	Master of Science (M.S.)
Degree Field	Computer Science
Degree Grantor	Oregon State University
Commencement Year	2022
Advisor	Chen, Lizhong
Committee Member	Bose, Bella Hamdaoui, Bechir
Academic Affiliation	Electrical Engineering and Computer Science
Rights Statement	In Copyright
Publisher	Oregon State University
Peer Reviewed	No
Language	English [eng]

Parents:

This work has no parents.

In Collection:

Thumbnail	Title	Date Uploaded	Visibility	Actions
	KaiZeng2021.pdf	2021-12-10	Public	Download