Topology-Aware Scheduling on Blue Waters with Proactive Queue Scanning and Migration-Based Job Placement.

Modern HPC systems, such as Blue Waters, have multidimensional torus topologies, which make it hard to achieve a high system utilization and a high scheduling efficiency. The low system utilization is majorly caused by system fragmentation, which includes both internal fragmentation due to convex prism shape requirement, and external fragmentation resulted from contiguous allocation strategy. The low scheduling efficiency comes from using a brute force search to find the free block with a matching shape for each job, which is highly time consuming. In this paper, we address the topology-aware scheduling problem on Blue Waters, with the objective of improving system utilization and scheduling efficiency. To improve scheduling efficiency, we propose an efficient free partition detection method. To improve system utilization, we propose a job scheduling strategy with proactive queue scanning and a migration-based job placement algorithm. Through extensive simulations of modeled trace data, we demonstrate that our approach improves the system utilization.