Development of grid job dispatcher based on lookahead scheduling
Abstract:
In this paper an implementation of the program complex GrAS (Grid Advanced Scheduler) for dispatching grid jobs is described. The main feature of GrAS is the usage of a Lookahead scheduling method. This method has the ability to build up a plan of resource allocations for some period of the time in the future. Because the plan relies on the accurate prediction of resource states the distribution of jobs may be done efficiently to avoid resource overloading. Also lookahead scheduling further to stage in job files in advance thereby minimizing idle time of resources. Program implementation of GrAS supports full cycle of job processing and is built on the basis of Globus Toolkit 3.
Keywords:
grid, distributed computing, dispatching, scheduling, resource reservation, cluster, quality of service
