Tux²: Distributed Graph Computation for Machine Learning

Authors: 

Wencong Xiao, Beihang University and Microsoft Research; Jilong Xue, Peking University and Microsoft Research; Youshan Miao, Microsoft Research; Zhen Li, Beihang University and Microsoft Research; Cheng Chen and Ming Wu, Microsoft Research; Wei Li, Beihang University; Lidong Zhou, Microsoft Research

Abstract: 

TUX2 is a new distributed graph engine that bridges graph computation and distributed machine learning. TUX2 inherits the benefits of an elegant graph computation model, efficient graph layout, and balanced parallelism to scale to billion-edge graphs; we extend and optimize it for distributed machine learning to support heterogeneity, a Stale Synchronous Parallel model, and a new MEGA (Mini-batch, Exchange, GlobalSync, and Apply) model.

We have developed a set of representative distributed machine learning algorithms in TUX2, covering both supervised and unsupervised learning. Compared to implementations on distributed machine learning platforms, writing these algorithms in TUX2 takes only about 25% of the code: Our graph computation model hides the detailed management of data layout, partitioning, and parallelism from developers. Our extensive evaluation of TUX2, using large data sets with up to 64 billion edges, shows that TUX2 outperforms state-of-the-art distributed graph engines PowerGraph and PowerLyra by an order of magnitude, while beating two state-of-the-art distributed machine learning systems by at least 48%.

NSDI '17 Open Access Videos Sponsored by
King Abdullah University of Science and Technology (KAUST)

Open Access Media

USENIX is committed to Open Access to the research presented at our events. Papers and proceedings are freely available to everyone once the event begins. Any video, audio, and/or slides that are posted after the event are also free and open to everyone. Support USENIX and our commitment to Open Access.

Presentation Video

Download Video

Presentation Audio

BibTeX
@inproceedings {201566,
author = {Wencong Xiao and Jilong Xue and Youshan Miao and Zhen Li and Cheng Chen and Ming Wu and Wei Li and Lidong Zhou},
title = {Tux{\texttwosuperior}: Distributed Graph Computation for Machine Learning},
booktitle = {14th {USENIX} Symposium on Networked Systems Design and Implementation ({NSDI} 17)},
year = {2017},
isbn = {978-1-931971-37-9},
address = {Boston, MA},
pages = {669--682},
url = {https://www.usenix.org/conference/nsdi17/technical-sessions/presentation/xiao},
publisher = {{USENIX} Association},
}