Many HPC workloads are now composed of multiple components running side by side on a node. These different parts have differing behaviors and resources usage which can lead to performance interference, from increased operating system noise to shared cache thrashing. One solution for these interference issues is to partition the hardware resources between components, using a wide range of advanced software and hardware mechanisms available on recent architectures.  In this talk, we will present cuttr, a resource partitioning tool under development to provide a unified and simplified access to these partitioning facilities. Early results on a CPU and cache partitioning policy used to improve a libEnsemble workload will be presented.

Alexis Janon is a PhD student at Inria Grenoble, working on low-level system software for resource management and performance interference mitigation.