Queue Prediction and Virtualized Scheduling Abstractions for NSF Batch-scheduled Cyberinfrastructure

University Of California-Santa Barbara, Santa Barbara CA

Investigators

Abstract

This project proposes to address and reduce the productivity-robbing uncertainty associated with queuing delay experienced by production batch queue users by providing a new and more accurate queue delay prediction services based on the QBETS system batch queue prediction system now available for the TeraGrid, developing and deploying a Virtual Advanced Reservation System (VARQ) that allows users to make probabilistic advanced reservations, and developing and deploying a Virtual Co-scheduling System (CO-VARQ) that combines probabilistic advanced reservations both to ?boost? success likelihood, and to enable co-scheduling. QBETS allows users to predict bounds on the delay individual jobs will experience while VARQ and COVARQ permit users to reserve in advance a specific future time slot for each job.