University of Houston
Department of Computer Science
In partial fulfillment of the Requirements for the Degree of
Master of Science
Shreenivasa Venkataramaiah
will defend his thesis
Performance Prediction for Distributed Applications
with Simple CPU and Network Sharing
Abstract
Efficient scheduling of any distributed application on a shared heterogeneous network requires a complete understanding of the application’s behavior along with the knowledge of current network conditions. Application performance varies depending on the available resources on the network. This variance is a characteristic of the application and estimation of the performance of the application under given network resource availability condition is of central importance for the problem of application scheduling and resource selection. This thesis addresses this problem of application performance modeling. The basic method employed in the characterization process is to perform fine-grain monitoring of the application execution on a dedicated cluster. We measure the different aspects of application execution, specifically CPU and communication behavior, through microscopic monitoring and we identify its pattern of resource usage and performance variance. We characterize the application by conducting system level measurements and avoid going through the code structure or the algorithm of the application. We discuss the characterization/predictor functions we have developed that would predict the application performance under given network conditions. Finally we validate these functions by experimenting with the NAS benchmarks. We run these benchmarks in network-sharing conditions and show that our predicted values are close to the actual ones from the experiments.
Date: Wednesday, August 28, 2002
Time: 1:30 PM
Place: 550-PGH
Faculty, students, and the general public are invited.
Thesis Advisor: Dr. Jaspal Subhlok