Recent advances in wide-area networking make feasible the execution of applications computers constructed from geographically distributed resources, or distributed systems. Two critical components for ease of use of such systems are methods for locating and scheduling the resources. These issues are complicated by the fact that the candidate set of resources is large, in the range of thousands of resources, and each resource may have its own, unique scheduling policy. Finding an ``optimal'' (or at least adequate) subset of resources that meet application requirements and are simultaneously available is the focus of this project.
Our work focuses on the basic issues related to how to represent an application requirements and how to co-schedule the various resources to satisfy the application requirements. Currently, we are performing extensive simulation studies and large-scale experiments to develop a method for predicting the time that a job must wait in a queue prior to execution. Once developed, the prediction methodology will be incorporated into a tool to simultaneous schedule resources, or co-schedule resources, to satisfy application requirements.
The following significant results were obtained this past year in this project.
Accomplishment 1
Conducted initial experiments that demonstrated the system characteristics necessary for reducing the error with predicting the queue wait time for parallel applications.