Beyond Shortest-Paths: A Benchmark for Reinforcement Learning on Traffic Engineering
Primary Area: datasets and benchmarks
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Keywords: Traffic Engineering, Routing Optimization, Multi-Agent Reinforcement Learning, Benchmark, Framework, Computer Networks
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
TL;DR: We view Distributed Traffic Engineering as a sequential decision making problem, provide a versatile learning framework and two promising routing optimization policies trained using swarm reinforcement learning.
Abstract: Selecting efficient routes for data packets is an essential task in computer networking. Given the dynamic of today’s network traffic, the optimal route varies greatly with the current network state. Despite the wealth of existing techniques, Traffic Engineering in networks with changing conditions is still a largely unsolved problem. Recent work aims at replacing Traffic Engineering heuristics with Reinforcement Learning, but does not provide a reference framework for training and evaluating under realistic network conditions in a reproducible manner. We fill this gap by casting distributed Traffic Engineering as a Swarm Markov Decision Process, and introducing a training and evaluation framework powered by a faithful network simulation engine that implements it. We show the effectiveness and versatility of our framework on a variety of scenarios, including ones where the agents outperform popular shortest-path routing algorithms.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 3616
Loading