Job Submission

How do I submit a job to the P1 DTU Cluster?

This cluster uses the IBM Platform LSF (Load Sharing Facility) as its workload management platform and job scheduler.

Static

static.sh

#!/bin/bash
#BSUB -J my_job_name
#BSUB -q p1
#BSUB -o %J.out
#BSUB -e %J.err
#BSUB -n 8
#BSUB -gpu "num=2:mode=exclusive_process"
#BSUB -R "rusage[mem=4GB]"
#BSUB -R "span[hosts=1]"
#BSUB -W 01:00

# Load the cuda module
module load cuda/11.6

# Check the GPU status
nvidia-smi

# Run the deviceQuery program
/appl/cuda/11.6.0/samples/bin/x86_64/linux/release/deviceQuery

bsub < static.sh

Dynamic

dynamic.sh

#!/bin/bash

set -euo pipefail  # safety flags

# Read scheduler configuration from environment variables (dynamic with sensible defaults)
: "${SCHEDULER_JOB_NAME:?Must provide SCHEDULER_JOB_NAME}"
: "${SCHEDULER_EMAIL:?Must provide SCHEDULER_EMAIL}"
: "${SCHEDULER_CORES:=4}"
: "${SCHEDULER_WALLTIME:=01:00}"
: "${SCHEDULER_QUEUE:=p1}"
: "${SCHEDULER_MEMORY:=4GB}"
: "${SCHEDULER_GPUS:=1}"

# Executed on node: Embedded job script using heredoc
JOB_SCRIPT=$(cat <<'EOF'
#!/bin/bash
set -euxo pipefail

# Load the cuda module
module load cuda/11.6

# Check the GPU status
nvidia-smi

# Run the deviceQuery program
/appl/cuda/11.6.0/samples/bin/x86_64/linux/release/deviceQuery
EOF
)

# Executed on host: Submit the job using CLI arguments
bsub \
  -J "${SCHEDULER_JOB_NAME}" \
  -q "${SCHEDULER_QUEUE}" \
  -o "%J.out" \
  -e "%J.err" \
  -n "${SCHEDULER_CORES}" \
  -gpu "num=${SCHEDULER_GPUS}:mode=exclusive_process" \
  -R "rusage[mem=${SCHEDULER_MEMORY}]" \
  -R "span[hosts=1]" \
  -W "${SCHEDULER_WALLTIME}" \
  <<< "$JOB_SCRIPT"

# Make the script executable
chmod +x dynamic.sh

# Source the .env file (optional)
source .env

# Or override explicitly
export SCHEDULER_JOB_NAME="my_job_name"
export SCHEDULER_CORES=8
export SCHEDULER_MEMORY=8GB
export SCHEDULER_GPUS=2
export SCHEDULER_WALLTIME=02:00
export SCHEDULER_QUEUE=p1

# Execute the script
./dynamic.sh

How do I submit a job to the P1 NGC Cluster?

This cluster uses the PBS (Portable Batch System) as its workload management platform and job scheduler.

static.job

#!/bin/bash
#PBS -A my_project_or_account_name
#PBS -N my_job_name
#PBS -l nodes=1:ppn=6:gpus=1
#PBS -l mem=5gb
#PBS -l walltime=3:00:00
#PBS -o %J.log
#PBS -j oe

cd $PBS_O_WORKDIR

# Load modules
module load cmake/3.22.2
module load miniconda3/4.12.0

# Activate Conda environment
conda activate MyEnv

# Call main entrypoint/script
python3 myscript.py

sbatch < static.job