MSRL: Distributed Reinforcement Learning with Dataflow Fragments

Huanzhou Zhu, Bo Zhao, Gang Chen, Weifeng Chen, Yijie Chen, Liang Shi, Yaodong Yang, Peter Pietzuch, Lei Chen

Research output: Chapter in Book/Report/Conference proceedingConference article in proceedingsScientificpeer-review

74 Downloads (Pure)

Abstract

A wide range of reinforcement learning (RL) algorithms have been proposed, in which agents learn from interactions with a simulated environment. Executing such RL training loops is computationally expensive, but current RL systems fail to support the training loops of different RL algorithms efficiently on GPU clusters: they either hard-code algorithm-specific strategies for parallelization and distribution; or they accelerate only parts of the computation on GPUs (e.g., DNN policy updates). We observe that current systems lack an abstraction that decouples the definition of an RL algorithm from its strategy for distributed execution.

We describe MSRL, a distributed RL training system that uses the new abstraction of a fragmented dataflow graph (FDG) to execute RL algorithms in a flexible way. An FDG is a heterogenous dataflow representation of an RL algorithm, which maps functions from the RL training loop to independent parallel dataflow fragments. Fragments account for the diverse nature of RL algorithms: each fragment can execute on a different device through a low-level dataflow implementation, e.g., an operator graph of a DNN engine, a CUDA GPU kernel, or a multi-threaded CPU process. At deployment time, a distribution policy governs how fragments are mapped to devices, without requiring changes to the RL algorithm implementation. Our experiments show that MSRL exposes trade-offs between different execution strategies, while surpassing the performance of existing RL systems with fixed execution strategies.
Original languageEnglish
Title of host publicationProceedings of the 2023 USENIX Annual Technical Conference
PublisherUSENIX -The Advanced Computing Systems Association
Pages977-993
ISBN (Electronic)978-1-939133-35-9
Publication statusPublished - 2023
MoE publication typeA4 Conference publication
EventUSENIX Annual Technical Conference - Boston, United States
Duration: 10 Jul 202312 Jul 2023
https://www.usenix.org/conference/atc23

Conference

ConferenceUSENIX Annual Technical Conference
Abbreviated titleATC
Country/TerritoryUnited States
CityBoston
Period10/07/202312/07/2023
Internet address

Fingerprint

Dive into the research topics of 'MSRL: Distributed Reinforcement Learning with Dataflow Fragments'. Together they form a unique fingerprint.

Cite this