Demystifying graph processing frameworks and benchmarks
Demystifying graph processing frameworks and benchmarks
复制标题
DOI：
10.1007/s11432-019-2807-4
复制
发表时间：
2020-06
期刊：
影响因子：
--
通讯作者：
Junyong Deng;Qinzhe Wu;Xiaoyan Wu;Shuang Song;Joseph Dean;L. John
中科院分区：
文献类型：
--
作者：
Junyong Deng;Qinzhe Wu;Xiaoyan Wu;Shuang Song;Joseph Dean;L. John
亲爱的编辑，由于从大数据中提取信息的需求不断增加，图算法变得非常重要[1]。由于大量数据和不规则的通信模式[2]，有效应用图算法是一项具有挑战性的任务。人们做出了大量努力来使用新颖的硬件设计来提高图形处理的性能，并通过实现各种框架来促进应用程序开发。此类框架的引入导致许多应用程序是通过不同的现有版本来实现的。例如，在研究中研究的四个框架/基准套件（即 GraphMat、图算法平台（GAP）、GraphBIG 和 Graph500）中，已经有六种广度优先搜索（BFS）实现、六种单源最短路径（SSSP）算法实现、五种三角形计数（TC）实现、四种 PageRank（PR）实现以及许多其他实现。 共同的应用。一般来说，研究和设计界并不总是完全清楚各种实现的特征。
Dear editor, Graph algorithms have become important because of the increasing need to extract information from big data [1]. Due to huge volumes of data and irregular communication patterns [2], it is a challenging task to apply graph algorithms efficiently. Considerable efforts have been made to improve the performance of graph processing using novel hardware designs and to facilitate application development by implementing various frameworks. Introduction of such frameworks has led to the fact that many applications are implemented through different existing versions. For instance, within the four frameworks/benchmark suites (ie, GraphMat, Graph Algorithm Platform (GAP), GraphBIG, and Graph500) investigated in the study, there are already six implementations of the breadth first search (BFS), six implementations of the single source shortest path (SSSP) algorithm, five implementations of triangle counting (TC), four implementations of PageRank (PR), and many other applications in common. Generally, the characteristics of various implementations are not always completely clear to the research and design community.