Job scheduling in high-performance computing is one of the fields that has seen few changes despite decades of progress in supercomputing. The application of system control theory is about to change this state of affairs. High-performance computing centers face serious challenges when it comes to satisfying processor resource allocation for exotic ecosystems of users. Even though automatic control is related to computer science, this is, to the best of my knowledge, the first application of automatic control techniques in job scheduling for predictable (in terms of deadline fulfillment) high-performance computing.
High-performance computing users can no longer afford to wait until the submitted job finishes without errors. Life-saving answers must be obtained in real time. For instance, tornado monitoring systems use high-performance numerical simulations for weather prediction. In fire analysis and monitoring systems or during computer-aided surgeries, any delay of the job scheduler may influence the outcome of the life-saving procedure.
This work is still in the prototyping phase, but it has been successfully tested with real-world applications. While some things could have been implemented within a less complicated framework, the work has the undeniable merit of blazing a trail. Anyone interested in high-performance computing should read this paper. It is well written, transparent, and paves the way for a new trend in job scheduling. It also teaches a good lesson: even closely related fields such as system control theory and computer science can have a definitive impact on each other when they interact.