Generating Traces of Application Behavior Using Generative Adversarial Networks Open Access

Abstract

Generally, applications, benchmarks and proxy applications are used for performance analysis of high-performance computing systems. These system performance analysis methods can be challenging or difficult to use, and often application traces are used as workload substitutes. But collecting these traces can also be difficult or time consuming. Therefore, we study synthetic trace generation to synthesize application trace that are indistinguishable from real traces. We propose a machine learning based synthetic data generation method that utilizes temporal graph generative adversarial networks (TG-GANs). We consider communication traces as temporal directed graphs with edge attributes and adjust TG-GAN to generate synthetic data. We use real traces as inputs and generate synthetic ones in some selected representative time windows and evaluate the quality of synthetic data using both quantitative metrics and visualizations. Visualization and quantitative results show that TG-GAN has the potential to generate high-quality synthetic traces but also has some limitations.

Table of Contents

1 Introduction 1

1.1 BackgroundandMotivation ................... 1

1.2 SyntheticTraceGeneration ................... 3

1.3 ContributionandStructure ................... 4

2 Background and Related Work 6

2.1 DataGeneration ......................... 6

2.2 GenerativeAdversarialNetworks ................ 8

2.3 Temporal Graph Generative Adversarial Network . . . . . . . 9

3 Approach 11

3.1 Applications............................ 11

3.2 ApplicationTrace......................... 12

3.3 Workflow ............................. 13

3.3.1 DataPreprocessing.................... 13

3.3.2 AdjustedTG-GAN .................... 16

3.3.3 EvaluationMethods ................... 17

4 Results and Analysis 20

4.1 TopologyandTemporalFeatures ................ 20

4.2 CommunicationIntensity..................... 25

4.3 SyntheticGenerationTime.................... 27

5 Summary and Future Work 30

5.1 Summary ............................. 30

5.2 FutureWork............................ 31

Appendix A - The Complete Set of Visualization Results 32

About this Master's Thesis

Rights statement

• Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.

School	Laney Graduate School
Department	Computer Science and Informatics
Degree	M.S.
Submission	Master's Thesis
Language	English
Research Field	Computer Science
Keyword	machine learning synthetic data generation generative adversarial networks
Committee Chair / Thesis Advisor	Arnold Dorian, Emory University
Committee Members	Vaidy Sunderam, Emory University Liang Zhao, Emory University

Last modified

Primary PDF

Thumbnail	Title	Date Uploaded	Actions
	Generating Traces of Application Behavior Using Generative Adversarial Networks ()	2021-05-06 17:00:02 -0400	Press to Select an action Download

Supplemental Files

Thumbnail	Title	Date Uploaded	Actions