Iker Martín Álvarez

Abstract:

High-Performance Computing (HPC) systems are growing increasingly complex, with diverse workloads competing for limited computational resources. Static resource allocation often leads to inefficiencies, including resource underutilization and increased job wait times, which hinder overall system performance and scalability. As workloads become more dynamic, there is a growing need for adaptive resource management strategies that can respond to changing demands in real-time.
Dynamic Resource Management (DRM) addresses these challenges by enabling jobs to adjust their resource allocation at runtime, optimizing system throughput and efficiency. 
This talk will provide an overview of dynamic resource allocation, exploring various reconfiguration techniques such as process management and data redistribution. Drawing from our own research, and collaborations, we will demonstrate how dynamic resource management can enhance workload efficiency, system utilization and reduce job completion time. Additionally, we will discuss the critical role of clear and interoperable user interfaces in making these techniques practical for adoption in production HPC environments.