Use this URL to cite or link to this record in EThOS:
Title: Adaptive performance control for scientific coupled models in heterogeneous and dynamic distributed environments
Author: Hussein, Mohamed Khamiss
ISNI:       0000 0001 3585 2340
Awarding Body: University of Manchester
Current Institution: University of Manchester
Date of Award: 2008
Availability of Full Text:
Access from EThOS:
The computational Grid environment is heterogeneous and has a highly dynamic nature. Consequently, applications executing in a Grid environment need to be controlled according to changes in the load conditions of the resources. This thesis proposes a novel adaptive performance control strategy. Its goal is to provide a reasonable execution time for an individual distributed application executing in a heterogeneous and dynamic distributed environment. The proposed adaptive performance control strategy employs distinct statistical prediction techniques, namely regression analysis and two different time series techniques, differing in complexity. The time series techniques are used to smooth the monitored execution times and to detect performance degradation of the distributed components of the application, as well as to provide short-term and long-term predictions for the distributed components of the application on their current resources. Regression analysis is used to provide predictions for the component behaviours on the available resources using a database of previous execution times. Using the above prediction techniques as a basis, the adaptive strategy reacts to changing load conditions on the resources by instigating a search process for a better mapping of the application components onto a subset of the allocated resources. The thesis proposes low cost search heuristics. The search heuristics take into account the load conditions on the resources, the cost of communication and the cost of migration. Experimental evaluation shows that the adaptive performance control strategy can yield significant performance improvement, saving up to 80% of the overall execution time in the reported experiments.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available