Applying reinforcement learning to scheduling strategies in an actual grid environment
International Journal of High Performance Systems Architecture
Hi-index | 0.00 |
One of the challenges in Grid computing research is to provide ameans to automatically submit, manage, and monitor applicationswhose main characteristic is to be composed of a large number oftasks. The large number of explicit tasks, generally placed on acentralized job queue, can cause several problems: (1) they canquickly exhaust the memory of the submission machine; (2) they candeteriorate the response time of the submission machine due tothese demanding too many open ports to manage remote execution ofeach of the tasks; (3) they may cause network traffic congestion ifall tasks try to transfer input and/or output files across thenetwork at the same time; (4) they make it impossible for the userto follow execution progress without an automatic tool orinterface; (5) they may depend on fault-tolerance mechanismsimplemented at application level to ensure that all tasks terminatesuccessfully. In this work we present and validate a novelarchitectural model, GRAND (Grid Robust ApplicatioN Deployment),whose main objective is to deal with the submission of a largenumbers of tasks. Copyright © 2006 John Wiley & Sons,Ltd.