Incorporating Job Predictions into the SEAGrid Science Gateway

Abstract

This paper describes the process of incorporating predictions of job queue wait times and run times into a Science Gateway. Science Gateways that integrate multiple resources can use predictions of queue wait times and run times to advice users when they choose where a job is executed or in an automated resource selection process. These predictions are also critical in executing workflows were it isn't feasible to have users specify where each task executes and the workflow management system therefore has to perform resource selection programmatically. SEAGrid science gateway has partly integrated the estimation of wait time prediction based on Karnak prediction service and is in the process of extending this to run time prediction.