Multisite Task Scheduling on Distributed Computing Grid Weizhe Zhang 1 , Hongli Zhang 1 , Hui He 2 , Mingzeng Hu 1 1 School of Computer Science and Technology, Harbin Institute of Technology, P.R.China {zwz, zhl, mzh}@pact518.hit.edu.cn http://pact518.hit.edu.cn/index.html 2 Network Information Center, Harbin Institute of Technology, P.R.China hehui@hit.edu.cn Abstract. Multisite task scheduling plays more and more important role in the grid computing as the WAN becomes faster and faster. Through the devel- opment of a three-level architecture of the distributed computing grid model and a grid schedule model, a scalable environment for multisite task schedul- ing is put forward. Then, a multisite Distributed Scheduling Server is designed and its prototype is implemented. A heuristic strategy, Clustering-based Grid Resource Selection algorithm, is described. Experiment indicates the sched- uler and the algorithm are effective. 1 Introduction Grid computing refers to the coordinated and secured sharing of computing re- sources across different administrative domains, aiming to solve the large-scale em- barrassing problems such as fluid dynamics, weather modeling, nuclear simulation and molecular modeling ….etc. Currently, computational grid can be classified into distributed computing grid and high-throughput computing grid [1]. A distributed supercomputing grid executes the application in parallel on multiple machines to reduce the completion time of a job. A high-throughput grid increases the comple- tion rate of a stream of jobs. Task scheduling is necessary and important to achieve less running time and higher throughput. Traditionally, the definition of task scheduling is the assignment of start and end times to a set of tasks to some certain resources, subject to certain constraints. How- ever, computing grid involves so many resources over multiple administrative do- mains that resources should be selected carefully in order to provide the best Qos. Thus, the traditional scheduling model based on static resources can not satisfy the large-scale dynamic resources requirement of the grid computing. In this paper, a new scheduling model oriented to the distributed computing grid is put forward. In the new scheduling model, the resource selection phrase plays an important role. Normally, resource selection algorithms can be classified into single-site and multi-site resource selection algorithms. Currently, most of the scheduler systems adopt the single-site resource selection algorithm such as Matchmaker/Class Ad