异构任务图在网格环境中的调度实现  

Heterogeneous task graph scheduling in grid environment

作　　者：华安 曾国荪[1,2] 乔伟光[1,2] 张飞[1,2] 

机构地区：[1]同济大学计算机科学及技术系 [2]国家高性能计算机工程技术中心同济分中心,上海200092

出　　处：《燕山大学学报》2005年第4期360-365,共6页Journal of Yanshan University

基　　金：国家自然科学基金项目(No.60173026);上海科委重大项目(No.03DZ15029);上海高校网格技术E-研究院资助项目(No.200301-1)

摘　　要：并行分布计算环境的任务调度一直是并行计算研究的热点问题,网格技术的发展则对任务调度提出了新的要求。本文采用带参数的有向超图表达异构任务图,为具有一般性、异构性、分布性、动态性等特点的任务请求提供了通用描述方法。通过设计网格异构任务图描述语言GHDL,实现了网格任务请求的统一封装以及任务请求在异构环境下的计算机理解和识别。通过设计自适应调度策略库,为分布异构动态网格环境下的任务调度提供有效解决方案,实现了在考虑网格任务间优先约束与数据关联的同时,采取异构宿主机识别、异构客户端自动创建等方式开发利用环境异构性与任务异构性,采取实时网格状态信息采集与任务迁移屏蔽了资源动态性。异构任务图在网格环境中的调度效果由树生长的仿真实验进行了展示。Task scheduling under parallel and distributed environment has been the research focus in the area of parallel computing, while new scheduling requirements are brought forward with the development of grid technology. In this paper, a directed hyper graph with parameters is proposed to describe heterogeneous task graph (HTG), which provides a common description mechanism for general, heterogeneous, distributed and dynamic task requests. With the design of GHDL (Grid HTG Definition Language), grid task requests are uniformly encapsulated, recognized and processed by heterogeneous computers. By means of self-adaptive scheduling strategy library, an effective solution for grid task scheduling in grid is provided, involving priority restriction and data correlation. Furthermore, the heterogeneity of environment and task is exploited through heterogeneous host recognition and au- tomatic client construction. Dynamic problem ofresource issolvedbyruntimeresource status monitoringand task migration. Simul ation experiment of trees in grid environment shows very promising results.

关 键 词：网格计算 有向超图 异构任务 并行调度 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]