A taskgraph clustering algorithm based on an attraction metric between tasks

CompEuro 1992 Proceedings Computer Systems and Software Engineering Pub Date : 1992-05-04 DOI:10.1109/CMPEUR.1992.218483

J. Opsommer

引用次数: 3

Abstract

Task granularity is a critical issue in the partitioning of a parallel program. To adjust automatically the task granularity for the target system, the solution used for the grain-size problem is the bottom-up approach: first the program is partitioned into elementary operations and mathematical functions and then several operations are combined into larger tasks. The conglomeration of tasks is controlled by attraction values: the attraction between two tasks is proportional to the benefit of aggregating the two tasks. The clustering heuristic is embedded in the definition of the attraction between tasks as only tasks with an attraction value higher than a certain threshold are conglomerated. It is assumed that the task graph structure and the task lengths are known at compile time. This information is used by the clustering algorithm. The algorithm defines an attraction between two tasks and then conglomerates the tasks for which the attraction is larger than a given threshold.<>

查看原文

微信好友朋友圈 QQ好友复制链接

本刊更多论文

一种基于任务间吸引力度量的任务图聚类算法

任务粒度是并行程序划分中的一个关键问题。为了自动调整目标系统的任务粒度，粒度问题的解决方案是自底向上的方法:首先将程序划分为初等运算和数学函数，然后将多个运算组合成更大的任务。任务的聚集受吸引力值的控制:两个任务之间的吸引力与两个任务聚集的利益成正比。聚类启发式嵌入到任务间吸引力的定义中，只有吸引力值高于某一阈值的任务才会被合并。假设任务图结构和任务长度在编译时是已知的。聚类算法使用这些信息。该算法定义两个任务之间的吸引力，然后合并吸引力大于给定阈值的任务

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文去求助

来源期刊

CompEuro 1992 Proceedings Computer Systems and Software Engineering

自引率

0.00%

发文量

期刊最新文献

Neural clustering algorithms for classification and pre-placement of VLSI cells General-to-specific learning of Horn clauses from positive examples Minimization of NAND circuits by rewriting-rules heuristic A generalized stochastic Petri net model of Multibus II Activation of connections to accelerate the learning in recurrent back-propagation