Main menu
You are here
Submitting jobs
Fri, 08/13/2021 - 13:01 — amitoj
Submitting Jobs to the Knights Landing Cluster
Submitting jobs to the Knights Landing cluster is similar to submitting to any other cluster at JLab. Currently one must submit to the phi queue:
$ sbatch -p phi -N noOfNodes -A slurmAccount job_script.sh
Job Size
Request the number of nodes needed:
# Ask for 1 node $ sbatch -p phi -N 1 -A slurmAccount ./job_script.sh
# Ask for 2 nodes $ sbatch -p phi -N 2 -A slurmAccount ./job_script.sh
MCDRAM and Cluster Mode Selection
One can ask for specific MCDRAM and cluster modes by specifying node-tags as usual. Initially we support cache-quadrant and flat-quadrant modes (see the system overview). The available tags are:
cache
- Cache MCDRAM - all or a part of the HBM is used as cache. No work required to use, but may have lower performance than flat mode in case of frequent misses in HBM as cache.
flat
-- Flat MCDRAM - the entirety of the HBM is used as addressable memory. May offer better performance than cache mode, but requires modifications of the code and/or execution environment.
equal
-- Flat MCDRAM - a portion of the HBM is used as addressable memory and the rest used as cache. Benefit of both Flat mode and Cache mode, but smaller sizes for each.
So for example to request 4 nodes, specifically with flat mode settings one would do
$ sbatch -p phi -N 4 -C flat -A slurmAccount ./job_script.sh
Other modes and tags may be supported in the future depending on demand. The default mode, if no tags are specified is next available node(s) in cache mode.