Predicting Available Computer Resources in Supercomputers

摘要

To maximize efficiency, the workload management system of a supercomputer must be able to process a multitude of job requests at a given time. The application of data processing techniques shows it is possible to predict node availability. The high error and variance of such predictions can be eliminated through the utilization of machine learning methods. Implementation of accurate predictions leads to more productive supercomputing systems, accelerating the pace of groundbreaking scientific discovery.