Adaptive Scheduling System for Grid Enviroment
Graduate School of Engineering & Technology, Korea University
Dept. of Electronics and Computer Engineering
on-line Adaptive Scheduling System Grid Enviroment;
- 원문 URL
In this thesis, we first classify scheduling algorithms into on-line, batch, QoS guided, workflow according to some criteria, and then describe several prediction algorithms: PACE, NWS, Bricks, Titan systems. Then we propose an adaptive scheduling system architecture which consists of three main components: Scheduler, Actuator and Metadata Bookkeeper. Scheduler component provides various scheduling algorithms for on-line mode, batch mode, QoS guided, and workflow. Actuator component transfers the related the files to the allocated resources, and execute them. Metadata Bookkeeper is used to store meta data which are obtained from the prediction model, and provide the scheduler component with the meta data which can be efficiently exploited in the scheduling algorithms. The advantage of our system is to select one of the diverse scheduling algorithms according to the property of the job as well as the meta data obtained from prediction model, thus providing adaptivity and performance improvement.