Abstract
dc:descriptionRecent networking research has identified that data-driven congestion control (CC) can be more efficient than traditional CC in data centers (DCs). Deep reinforcement learning (RL), in particular, has the potential to learn optimal network policies. However, RL suffers from instability and over-fitting, deficiencies which so far render it unacceptable for use in DC networks. We analyze the requirements for data-driven policies to succeed in the DC context. And, we present a new emulator, Iroko, which supports different network topologies, DC traffic engineering algorithms, and deployment scenarios. Iroko interfaces with the OpenAI gym toolkit, which allows for fast and fair evaluation of RL against traditional algorithms under equal conditions. We present initial benchmarks of three deep RL algorithms against TCP New Vegas and DCTCP. The results show that out-of-the-box algorithms are able to learn a CC policy with comparative performance to TCP in our emulator. We make Iroko open-source and publicly available: https://github.com/dcgym/iroko.
Degree
thesis:*- Name thesis:degree_name
- Master of Science - MSc
- Level thesis:degree_level
- master's
- Discipline thesis:degree_discipline
- Computer Science
- Grantor dc:publisher
- University of British Columbia
- Year dc:date
- 2019
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Ruffy Varga, Fabian
Rights
dc:rights- Statement dc:rights
-
- Attribution-NonCommercial 4.0 International
- Language dc:language
- eng
Identifiers
dc:identifier.*- Handle dc:identifier
- http://hdl.handle.net/2429/69869