Oh Han Kang , Sang Sung Kang , Hee Heon Song

The KIPS Transactions:PartA, Vol. 16, No. 1, pp. 9-16, Feb. 2009
10.3745/KIPSTA.2009.16.1.9,   PDF Download:

Abstract

Because Grid system consists of heterogenous computing resources, which are distributed on a wide scale, it is impossible to efficiently execute applications with scheduling algorithms of a conventional parallel system that, in contrast, aim at homogeneous and controllable resources. To suggest an algorithm that can fully reflect the characteristics of a grid system, our research is focused on examining the type of information used in current scheduling algorithms and consequently, deriving factors that could develop algorithms further. The results from the analysis of these algorithms not only show that static information of resources such as capacity or the number of processors can facilitate the scheduling algorithms but also verified a decrease in efficiency in case of utilizing real time load information of resources due to the intrinsic characteristics of a grid system relatively long computing time, and the need for the means to evade unfeasible resources or ones with slow processing time. In this paper, we propose a new algorithm, which is revised to reflect static information in the logic of WQR (Workqueue Replication) algorithms and show that it provides better performance than the one used in the existing method through simulation.

Statistics
Cite this article