Data-driven versus Topology-driven Irregular Computations on GPUs

被引:68
作者
Nasre, Rupesh [1 ]
Burtscher, Martin [2 ]
Pingali, Keshav [1 ]
机构
[1] Univ Texas Austin, Austin, TX 78712 USA
[2] Texas State Univ, San Marcos, TX USA
来源
IEEE 27TH INTERNATIONAL PARALLEL AND DISTRIBUTED PROCESSING SYMPOSIUM (IPDPS 2013) | 2013年
基金
美国国家科学基金会;
关键词
irregular algorithms; data-driven; topology-driven; algorithmic properties; GPGPU; GRAPH ALGORITHMS; PARALLELISM;
D O I
10.1109/IPDPS.2013.28
中图分类号
TP301 [理论、方法];
学科分类号
081202 ;
摘要
Irregular algorithms are algorithms with complex main data structures such as directed and undirected graphs, trees, etc. A useful abstraction for many irregular algorithms is its operator formulation in which the algorithm is viewed as the iterated application of an operator to certain nodes, called active nodes, in the graph. Each operator application, called an activity, usually touches only a small part of the overall graph, so non-overlapping activities can be performed in parallel. In topology-driven implementations, all nodes are assumed to be active so the operator is applied everywhere in the graph even if there is no work to do at some nodes. In contrast, in data-driven implementations the operator is applied only to nodes at which there might be work to do. Multicore implementations of irregular algorithms are usually data-driven because current multicores only support small numbers of threads and work-efficiency is important. Conversely, many irregular GPU implementations use a topology-driven approach because work inefficiency can be counterbalanced by the large number of GPU threads. In this paper, we study data-driven and topology-driven implementations of six important graph algorithms on GPUs. Our goal is to understand the tradeoffs between these implementations and how to optimize them. We find that data-driven versions are generally faster and scale better despite the cost of maintaining a worklist. However, topology-driven versions can be superior when certain algorithmic properties are exploited to optimize the implementation. These results led us to devise hybrid approaches that combine the two techniques and outperform both of them.
引用
收藏
页码:463 / 474
页数:12
相关论文
共 50 条
  • [31] A Framework for Data-Driven Automata Design
    Zhang, Yuanrui
    Chen, Yixiang
    Ma, Yujing
    REQUIREMENTS ENGINEERING IN THE BIG DATA ERA, 2015, 558 : 33 - 47
  • [32] On the Data-Driven Modeling of Reactive Extrusion
    Ibanez, Ruben
    Casteran, Fanny
    Argerich, Clara
    Ghnatios, Chady
    Hascoet, Nicolas
    Ammar, Amine
    Cassagnau, Philippe
    Chinesta, Francisco
    FLUIDS, 2020, 5 (02)
  • [33] Data-driven Approaches to Edge Caching
    Li, Guangyu
    Shen, Qiang
    Liu, Yong
    Cao, Houwei
    Han, Zifa
    Li, Feng
    Li, Jin
    PROCEEDINGS OF THE 2018 WORKSHOP ON NETWORKING FOR EMERGING APPLICATIONS AND TECHNOLOGIES (NEAT '18), 2018, : 8 - 14
  • [34] Precomputing data-driven tree animation
    Zhang, Long
    Zhang, Yubo
    Jiang, Zhongding
    Li, Luying
    Chen, Wei
    Peng, Qunsheng
    COMPUTER ANIMATION AND VIRTUAL WORLDS, 2007, 18 (4-5) : 371 - 382
  • [35] A data-driven method for dissipative thermomechanics
    Ruiz, D.
    Portillo, D.
    Romero, I
    IFAC PAPERSONLINE, 2021, 54 (19): : 315 - 320
  • [36] Data-driven deep density estimation
    Puchert, Patrik
    Hermosilla, Pedro
    Ritschel, Tobias
    Ropinski, Timo
    NEURAL COMPUTING & APPLICATIONS, 2021, 33 (23) : 16773 - 16807
  • [37] Efficient Data-Driven Network Functions
    Yao, Zhiyuan
    Desmouceaux, Yoann
    Cordero-Fuertes, Juan-Antonio
    Townsley, Mark
    Clausen, Thomas
    2022 30TH INTERNATIONAL SYMPOSIUM ON MODELING, ANALYSIS, AND SIMULATION OF COMPUTER AND TELECOMMUNICATION SYSTEMS, MASCOTS, 2022, : 152 - 159
  • [38] TOWARD DATA-DRIVEN FILTERS IN PARAVIEW
    Maharjan D.
    Zaspel P.
    Journal of Flow Visualization and Image Processing, 2022, 29 (03) : 55 - 72
  • [39] Fuzzy and Data-Driven Urban Crowds
    Toledo, Leonel
    Rivalcoba, Ivan
    Rudomin, Isaac
    COMPUTATIONAL SCIENCE - ICCS 2018, PT III, 2018, 10862 : 280 - 290
  • [40] Data-Driven Approach for Spellchecking and Autocorrection
    Toleu, Alymzhan
    Tolegen, Gulmira
    Mussabayev, Rustam
    Krassovitskiy, Alexander
    Ualiyeva, Irina
    SYMMETRY-BASEL, 2022, 14 (11):