부산대학교 도서관

Implementing efficient dynamic scheduling algorithms is a real challenge but a well designed algorithm can bring a significant performance improvement, regardless of unexpected events that may occur during execution. What is more, analyzing an application's needs and ensuring the most appropriate course of action in case of a delay or failure is bound to offer the best performance for the application in question. This paper presents three dynamic scheduling algorithms for workflows, implemented at application level. The application does satellite image processing, by describing a complex operation as a workflow of elementary operators. The scheduling process described by these algorithms doesn't control the resources directly so it is more natural to consider it closer to Grid applications. The algorithms are responsible for the management of tasks in workflow, such as managing the tasks for parallel execution, managing of data and correlation of events. To fulfill their functions, the scheduling algorithms need information coming from monitoring services available in the execution platform. The platform provides a series of libraries and services as well as a management and execution mechanism which will be used to test each algorithm's efficiency. The comparison between proposed dynamic algorithms and the platform's initial scheduling mechanism highlights the obtained improvements referring to the workflow execution time.