Smart resource management beyond compute nodes
Monday, May 30, 2022 5:30 PM to 6:30 PM · 1 hr. (Europe/Berlin)
Hall D - 2nd Floor
Exascale SystemsHPC Workflows
Information
As HPC environments evolve towards the Exascale, new challenges appear. In particular a wider heterogeneity regarding the resources is to be managed. In the “Good Old Time”, the term “resources” was often a shortcut for “CPUs, amount of RAM and number of compute nodes”, but this concept has changed a lot and is becoming more complex: a strong diversity of the compute resources, deep storage tiers with very different devices (from NVRAM to tapes), and also with extended network features allowing for optimisation (via adaptive routing, congestion management, QoS, …). This heterogeneity of the underlying system requires to be “smart” when allocating resources.
In this BoF, we will look into what it means to be “smart” from different points of view: from application level, to IO-dedicated systems and to the network layers. This BoF is about discussing and sharing amongst us the upcoming challenges we see in resources management, and also in discussing possible ways to tackle them.
Contributors:
In this BoF, we will look into what it means to be “smart” from different points of view: from application level, to IO-dedicated systems and to the network layers. This BoF is about discussing and sharing amongst us the upcoming challenges we see in resources management, and also in discussing possible ways to tackle them.
Contributors:
- Estela Suarez (Jülich Supercomputing Centre)
- Philippe DENIEL (CEA/DAM)
- Norbert Eicker (FZJ)
- Grégoire Pichon (ATOS)
- Pascale Bernier-Bruna (Atos)
- Maike Gilliot (CEA)
Format
On-site
