Reda Ammar, Ph.D.
Computer Science
and Engineering Department
University of Connecticut
Storrs, Connecticut
Date:
Wednesday April 25, 2007
Time:
Location: 322 Votey
Abstract
Scheduling a large number of high performance computing applications on cluster environment is a serious obstacle to achieving a good performance. This becomes more critical in real time systems. A cluster scheduler without enough knowledge of the state of the cluster and the scheduled tasks cannot adequately manage the cluster resources. Consequently, it may fragment the available processing power. This may cause rejection of some submitted applications due to tasks missing their deadlines. Few researchers have investigated the problem of scheduling real-time task graphs; however they have not provided mechanisms to satisfy the performance requirements while maximizing the processing power utilization. In this work, we focused on developing and evaluating scheduling techniques for real-time applications on cluster computing systems. These applications are represented as task graphs of different structures (tandem, fork join structure, conditional structure and loops) that arrive randomly to different nodes of the cluster. We have showed that using applications' task graphs give better results than treating each application as a single task unit.