Advisor

Banicescu, Ioana

Committee Member

Skjellum, Anthony

Committee Member

Bridges, Susan

Committee Member

Philip, Thomas

Committee Member

Haupt, Tomasz

Date of Degree

1-1-2008

Document Type

Dissertation - Open Access

Major

Computer Science and Engineering

Degree Name

Doctor of Philosophy

College

James Worth Bagley College of Engineering

Department

Department of Computer Science and Engineering

Abstract

Adaptive parallel applications that can change resources during execution, promise increased application performance and better system utilization. Furthermore, they open the opportunity for developing a new class of parallel applications driven by unpredictable data and events. The research issues in an adaptive parallel system are complex and interrelated. The nature and complexities of the relationships among these issues are not well researched and understood. Before developing adaptive applications or an infrastructure support for adaptive applications, these issues need to be investigated and studied in detail. One way of understanding and investigating these issues is by modeling and simulation. A model for adaptive parallel systems has been developed to enable the investigation of the impact of malleable workloads on its performance. The model can be used to determine how different model parameters impact the performance of the system and to determine the relationships among them Subsequently, a discrete event simulator has been developed to numerically simulate the model. Using the simulator, the impact of the variation in the number of malleable jobs in the workload, the flexibility, the negotiation cost, and the adaptation cost on system performance have been studied. The results and conclusions of these simulation experiments are presented in this dissertation. In general, the simulation results reveal that the performance improves with an increase in the number of malleable jobs in a workload, and that the performance saturates at a certain percentage of rigid to malleable jobs mix. A high percentage of malleable jobs is not necessary to achieve significant improvement in performance. The performance in general improves as the flexibility increases up to a certain point; then, it saturates. The negotiation cost impacts the performance, but not significantly. The number of negotiations for a given workload increases as number of malleable jobs increases up to a certain point, and then it decreases as number of malleable jobs increases further. The performance degrades as the application adaptation cost increases. The impact of the application adaptation cost on performance is much more significant compared to that of the negotiation cost.

URI

https://hdl.handle.net/11668/21269

Comments

Application Adaptation||Resource Negotiation||Resource Management System||Adaptive Parallel System||Malleable Application||Adaptive Application||Scheduling

Share

COinS