Work Stealing in a Shared Virtual-Memory Heterogeneous Environment A Case Study with Betweenness Centrality

被引:1
|
作者
Che, Shuai [1 ]
Orr, Marc [1 ]
Gallmeier, Jonathan [1 ]
机构
[1] Adv Micro Devices Inc, Santa Clara, CA 95054 USA
关键词
Heterogeneous system architecture; work stealing; graph algorithms;
D O I
10.1145/3075564.3075567
中图分类号
TP301 [理论、方法];
学科分类号
081202 ;
摘要
This paper uses betweenness centrality as a case study to research efficient work stealing in a heterogeneous system environment. Betweenness centrality is an important algorithm in graph processing. It presents multiple-level parallelism and is an interesting problem to exploit various optimizations. We investigate queue-based work stealing to distribute its tasks across GPU compute units (CUs) and across the CPU and the GPU, which has not been done by prior work. In particular, we demonstrate how to leverage the new platform-atomic operations on AMD Accelerated Processing Units (APUs) to operate cross-device queues in a lock-free manner in shared virtual memory. To make the work stealing runtime and the application more efficient, we apply new architectural features, including atomic operations with different memory scopes and orderings for different synchronization scenarios. We implement our solution using heterogeneous system architecture (HSA). Our results show that betweenness centrality with CPU-GPU work stealing achieves an average of 15% (up to 30%) performance improvement over GPU-only execution for diverse graph inputs. Our work stealing solution can be applied widely to other applications too. Finally, we analyze important parameters critical for queuing and stealing.
引用
收藏
页码:164 / 173
页数:10
相关论文
共 50 条
  • [1] EXPERIMENTING WITH A SHARED VIRTUAL MEMORY ENVIRONMENT FOR HYPERCUBES
    AGARWALA, A
    DAS, CR
    JOURNAL OF PARALLEL AND DISTRIBUTED COMPUTING, 1995, 29 (02) : 228 - 235
  • [2] The Mathematical Model and the Problem of Optimal Partitioning of Shared Memory for Work-Stealing Deques
    Sokolov, Andrew
    Barkovsky, Eugene
    PARALLEL COMPUTING TECHNOLOGIES (PACT 2015), 2015, 9251 : 102 - 106
  • [3] Observations and Opportunities in Architecting Shared Virtual Memory for Heterogeneous Systems
    Vesely, Jan
    Basu, Arkaprava
    Oskin, Mark
    Loh, Gabriel H.
    Bhattacharjee, Abhishek
    2016 IEEE INTERNATIONAL SYMPOSIUM ON PERFORMANCE ANALYSIS OF SYSTEMS AND SOFTWARE ISPASS 2016, 2016, : 161 - 171
  • [4] Staccato: shared-memory work-stealing task scheduler with cache-aware memory management
    Kuchumov, Ruslan
    Sokolov, Andrey
    Korkhov, Vladimir
    INTERNATIONAL JOURNAL OF WEB AND GRID SERVICES, 2019, 15 (04) : 394 - 407
  • [5] The Models and Methods of Optimal Control of Three Work-Stealing Deques Located in a Shared Memory
    E. A. Aksenova
    E. A. Barkovsky
    A. V. Sokolov
    Lobachevskii Journal of Mathematics, 2019, 40 : 1763 - 1770
  • [6] SIAW: An Adaptive Idleness-Aware Work-Stealing Strategy on Shared Memory Machines
    Guan, Lei
    LU, Yu-tong
    Gao, Tao
    2016 INTERNATIONAL CONFERENCE ON INFORMATION SYSTEM AND ARTIFICIAL INTELLIGENCE (ISAI 2016), 2016, : 247 - 252
  • [7] The Models and Methods of Optimal Control of Three Work-Stealing Deques Located in a Shared Memory
    Aksenova, E. A.
    Barkovsky, E. A.
    Sokolov, A. V.
    LOBACHEVSKII JOURNAL OF MATHEMATICS, 2019, 40 (11) : 1763 - 1770
  • [8] Q-measures and betweenness centrality in a collaboration network: a case study of the field of informetrics
    Guns, Raf
    Liu, Yu Xian
    Mahbuba, Dilruba
    SCIENTOMETRICS, 2011, 87 (01) : 133 - 147
  • [9] Q-measures and betweenness centrality in a collaboration network: a case study of the field of informetrics
    Raf Guns
    Yu Xian Liu
    Dilruba Mahbuba
    Scientometrics, 2011, 87 : 133 - 147
  • [10] Staccato: Cache-Aware Work-Stealing Task Scheduler for Shared-Memory Systems
    Kuchumov, Ruslan
    Sokolov, Andrey
    Korkhov, Vladimir
    COMPUTATIONAL SCIENCE AND ITS APPLICATIONS - ICCSA 2018, PT IV, 2018, 10963 : 91 - 102