Adaptive scheduling of multiscale processes in dynamically reconfigurable networked computers
Date of Completion
January 2003
Keywords
Computer Science
Degree
Ph.D.
Abstract
Grid computing—also known as Metacomputing—is an abstraction by which clusters of loosely coupled computers in a distributed system can be treated as a single virtual machine. With the aid of Grid middleware, clusters of loosely coupled machines (quite possibly heterogeneous) can be treated as a single virtual system or a virtual organization ( VO). ^ The optimal exploitation of resources within the Grid figures largely in the scheduling world. However, Grid computational resources are distributed, heterogeneous, and dynamic which makes it difficult to schedule applications in a way that maximizes appropriate performance metrics. In addition, the scheduling problem in its most general form is known to be NP-complete. Therefore, many heuristics have been developed to generate adequate (but sub-optimal) schedulers. ^ This research work focuses on developing algorithms for adaptively scheduling applications in dynamic multi-user distributed Grid environment based on three primary factors: an adaptive hierarchical Grid Monitoring policy, an adaptive hierarchical Grid Resource Forecasting policy, and an Intelligent Hierarchical Grid Scheduling policy. ^ Our objective is to optimize the services requested by user/node in the federated clusters to meet some criteria of performance optimality, ( such as Mean Flow Time). ^
Recommended Citation
Kerasha, Mohamed A, "Adaptive scheduling of multiscale processes in dynamically reconfigurable networked computers" (2003). Doctoral Dissertations. AAI3101697.
https://digitalcommons.lib.uconn.edu/dissertations/AAI3101697