Euge: Effective Utilization of GPU Resources for Serving DNN-Based Video Analysis

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

4 Scopus citations

Abstract

Deep Neural Network (DNN) has been widely adopted in video analysis application. The computation involved in DNN is more efficient on GPUs than on CPUs. However, recent serving systems involve the low utilization of GPU, due to limited process parallelism and storage overhead of DNN model. We propose Euge, which introduces multi-process service (MPS) and model sharing technology to support effective utilization of GPU. With MPS technology, multiple processes overcome the obstacle of GPU context and execute DNN-based video analysis on one GPU in parallel. Furthermore, by sharing the DNN-based model among threads within a process, Euge reduces the GPU memory overhead. We implement Euge on Spark and demonstrate the performance of vehicle detection workload.

Original languageEnglish
Title of host publicationWeb and Big Data - 4th International Joint Conference, APWeb-WAIM 2020, Proceedings
EditorsXin Wang, Rui Zhang, Young-Koo Lee, Le Sun, Yang-Sae Moon
PublisherSpringer Science and Business Media Deutschland GmbH
Pages523-528
Number of pages6
ISBN (Print)9783030602895
DOIs
StatePublished - 2020
Event4th Asia-Pacific Web and Web-Age Information Management, Joint Conference on Web and Big Data, APWeb-WAIM 2020 - Tianjin, China
Duration: 18 Sep 202020 Sep 2020

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume12318 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference4th Asia-Pacific Web and Web-Age Information Management, Joint Conference on Web and Big Data, APWeb-WAIM 2020
Country/TerritoryChina
CityTianjin
Period18/09/2020/09/20

Keywords

  • DNN
  • GPU
  • MPS
  • Model sharing

Fingerprint

Dive into the research topics of 'Euge: Effective Utilization of GPU Resources for Serving DNN-Based Video Analysis'. Together they form a unique fingerprint.

Cite this