Multi-application Scheduling in Networks of Workstations and Clusters of Processors
IWCC '01 Proceedings of the NATO Advanced Research Workshop on Advanced Environments, Tools, and Applications for Cluster Computing-Revised Papers
Dynamic, capability-driven scheduling of DAG-based real-time jobs in heterogeneous clusters
International Journal of High Performance Computing and Networking
| Hi-index | 0.00 |
In this paper, we present a dynamic approach for constructing and scheduling parallel adaptive applications in heterogeneous multi-user environments (Networks of Workstations). Parallel adaptive applications have the property of varying their parallelism degree following the load fluctuation of the underlying environment. Our tool provides a programming facility that allows the application construction to avoid managing these complex problems and an allocation module responsible for running and scheduling application tasks. The allocation module handles also all problems related to the dynamic character of the application so that the user may not know at any time whether his application executes on one or dozens of workstations. The allocation module is completed by a scheduler which tries to make good mapping decisions and to adjust the mapping when the application reconfigures dynamically. The scheduling approach based on the dependency graphs model tries to minimize the execution time of the application by decreasing the parallelism loss situations in which some nodes allocated to the application are waiting for the work availability which must be generated by some slow nodes. This can be achieved by analyzing dynamically the depgraph structure and using the heterogeneity aspect. Encouraging results were obtained from experiments conducted on a parallel version of the Gaussian elimination application which is not well adapted to our environment. The advantages of our tool are mainly: the adaptive aspect which allows the parallel application to exploit idle cycles in a cluster of workstations and the programming facility which decreases the parallel programming complexity.