Adaptive Selection of Partition Size for Supercomputer Requests

In this paper, we show how application scheduling can be used to reduce the turn-around time of supercomputer jobs. Our approach focuses on the use of SA, an AppLeS application scheduler, to adaptively craft the request to be submitted to the supercomputer based on the current state of the system. We demonstrate that SA significantly improves a job’s turn-around time in a variety of scenarios. We also identify how the state of the system, the characteristics of the job, and the quality of the information made available to SA influence its performance.