Section: New Results

Topology Aware Resource Management

SLURM  [24] is a Resource and Job Management System, a middleware in charge of delivering computing power to applications in HPC systems. Our goal is to take in account in SLURM placement process hardware topology as well as application communication pattern. We have a new selection option for the cons_res plugin in SLURM. In this case the usually BestFit algorithm used to choose nodes is replaced by TreeMatch to find the best placement among the free nodes list in light of a given application communication matrix.

We updated this plugin based on SLURM 2.6.5 for last version SLURM 15.08. To decrease the overhead due to our algorithm we also implemented an alternative to use a subtree of the global topology. We ran experiments to compare these different solutions using our plugin with or without subtree and the current algorithm topology-aware in SLURM.