Simulation vs Actual Walltime Correction in a Real Production Resource-Constrained HPC
Aira Villapando, Jessi Christa Rubio · Practice and Experience in Advanced Research Computing · 2021
Today’s increase in computational resource demand requires robust job scheduling performance in shared HPC systems, which have varying workload characteristics. We applied a previous theoretical work on a walltime corrective scheduling to a full-scale production-level resource-constrained HPC cluster. We analyzed its performance by comparing the simulated and actual production implementation with various workload scenarios. The results showed a significant difference between theory and reality and thus rendered us substantial insights to adjust the scheduling policies accordingly for effective resource management that similar resource-constrained HPC clusters could benefit from.