Lab Home | Phone | Search | ||||||||
|
||||||||
Traditional static resource allocation in supercomputers (jobs retain a fixed set of resources) leads to inefficiencies. Resource adaptivity (jobs can change resources at runtime) significantly increases supercomputer efficiency. This talk will exploit Asynchronous Many-Task (AMT) programming, which is well suited for adaptivity due to its transparent resource management. An AMT runtime system dynamically assigns user-defined small tasks to workers to achieve load balancing and adapt to resource changes. We will discuss techniques for malleability and evolving capabilities that allow programs to dynamically change resources without interrupting computation. Automatic load detection heuristics determine when to start or terminate processes, which is particularly beneficial for unpredictable workloads. Practicality is demonstrated by adapting the GLB library. A generic communication interface allows interaction between programs and resource managers. Evaluations with a prototype resource manager show significant improvements in batch makespan, node utilization, and job turnaround time for both malleable and evolving jobs. Bio: Jonas is a dedicated computer scientist specializing in High Performance Computing. He received his Bachelor’s and Master’s degrees from the University of Kassel, Germany, where he also earned his Ph.D. in 2022. He is currently working as an substitute chair for the Software Engineering Group at the same university and is also writing his habilitation. Jonas' research interests include load balancing, fault tolerance, and resource adaptivity for Asynchronous Many-Task (AMT) systems. Recently, he has focused on resource adaptivity in general to optimize the efficient use of supercomputing resources. His work covers a broad spectrum, including the development of advanced job scheduling algorithms, the improvement of application programming using AMT systems, and the interaction between resource managers and jobs. Teams: Join the meeting now Host: Patrick Diehl (CCS-7) |