摘要
网格计算技术是继Internet计算之后出现的新兴研究领域。网格系统由异构的资源组成,一个好的任务调度方法可以充分利用网格系统的处理能力,减少任务的完成时间。根据目前网格系统的使用模式,提出了符合实际的用户任务形式,即任务由数据传输和计算两部分组成,计算在获得所有输入之后开始执行。多个这样的独立任务组成元任务,作为调度程序的最小执行单位。在实际应用中,元任务应该由数据密集型和计算密集型任务混合组成。考虑到数据传输和计算的比例关系对元任务完成的影响,提出一种新的调度算法TCR,通过提高计算资源的利用率以及任务间的并行度,减少元任务的完成时间。详细介绍了该算法,并通过模拟结果的对比验证了该算法的良好性能。
Grid computing technology is a new research field following the Internet computing technology. Grid system consists of heterogeneous resources, so the major challenge of task scheduling is effectively by using the whole grid power to minimize the task makespan. According to the modei of using grid system, one user task can be divided into two components, the data transfer part and computation part. Computation starts after receiving all the input data. The independent set of such tasks that is considered for scheduling is called a metatask, usually composed by data intensive and computation intensive tasks randomly. So the ratio of transfer and computation will give big impact to the makespan of metatask. A new algorithm named TCR is developed to minimize the makespan by increasing computation resources utilization and task parallelism. Finally, the TCR algorithm is illustrated by details with the simulation results comparison in order to verify its good performance.
出处
《计算机工程与设计》
CSCD
2003年第10期1-4,共4页
Computer Engineering and Design
基金
国家杰出青年科学基金资助项目(69925205)