Example Scripts¶

Complete before/after examples for common job types. Each example shows the full LSF script alongside its SLURM equivalent.

Simple job¶

LSFSLURM

#!/bin/bash
#BSUB -J simple_job
#BSUB -q normal
#BSUB -W 1:00
#BSUB -n 1
#BSUB -M 4000
#BSUB -o simple.%J.out
#BSUB -e simple.%J.err

module load R/4.3.0

Rscript my_analysis.R

#!/bin/bash
#SBATCH --job-name=simple_job
#SBATCH --partition=normal
#SBATCH --time=01:00:00
#SBATCH --ntasks=1
#SBATCH --mem=4G
#SBATCH --output=simple.%j.out
#SBATCH --error=simple.%j.err

module load R/4.3.0

Rscript my_analysis.R

Array job¶

LSFSLURM

#!/bin/bash
#BSUB -J "align[1-50]%10"
#BSUB -q normal
#BSUB -W 4:00
#BSUB -n 8
#BSUB -M 16000
#BSUB -o logs/align.%J.%I.out
#BSUB -e logs/align.%J.%I.err

module load samtools/1.17
module load bwa/0.7.17

SAMPLE=$(sed -n "${LSB_JOBINDEX}p" samples.txt)
R1=fastq/${SAMPLE}_R1.fastq.gz
R2=fastq/${SAMPLE}_R2.fastq.gz

bwa mem -t 8 ref/genome.fa $R1 $R2 \
    | samtools sort -@ 4 -o bam/${SAMPLE}.sorted.bam

samtools index bam/${SAMPLE}.sorted.bam

#!/bin/bash
#SBATCH --job-name=align
#SBATCH --partition=normal
#SBATCH --time=04:00:00
#SBATCH --cpus-per-task=8
#SBATCH --mem=16G
#SBATCH --array=1-50%10
#SBATCH --output=logs/align.%A.%a.out
#SBATCH --error=logs/align.%A.%a.err

module load samtools/1.17
module load bwa/0.7.17

SAMPLE=$(sed -n "${SLURM_ARRAY_TASK_ID}p" samples.txt)
R1=fastq/${SAMPLE}_R1.fastq.gz
R2=fastq/${SAMPLE}_R2.fastq.gz

bwa mem -t 8 ref/genome.fa $R1 $R2 \
    | samtools sort -@ 4 -o bam/${SAMPLE}.sorted.bam

samtools index bam/${SAMPLE}.sorted.bam

GPU job¶

LSFSLURM

#!/bin/bash
#BSUB -J gpu_train
#BSUB -q gpu
#BSUB -W 24:00
#BSUB -n 4
#BSUB -M 32000
#BSUB -R "select[ngpus_physical>0] rusage[ngpus_physical=1]"
#BSUB -o train.%J.out
#BSUB -e train.%J.err

module load cuda/11.8
module load anaconda3

conda activate ml_env

python train.py \
    --epochs 100 \
    --batch-size 64 \
    --gpus 1

#!/bin/bash
#SBATCH --job-name=gpu_train
#SBATCH --partition=gpu
#SBATCH --time=24:00:00
#SBATCH --cpus-per-task=4
#SBATCH --mem=32G
#SBATCH --gpus=1
#SBATCH --output=train.%j.out
#SBATCH --error=train.%j.err

module load cuda/11.8
module load anaconda3

conda activate ml_env

python train.py \
    --epochs 100 \
    --batch-size 64 \
    --gpus 1

Job with dependencies¶

LSFSLURM

# Step 1: Alignment
JOB1=$(bsub -J align -q normal -W 4:00 -n 8 -o align.%J.out < align.sh | grep -oP '\d+')

# Step 2: Sort (after align completes successfully)
JOB2=$(bsub -J sort -q normal -W 2:00 -n 4 -w "done($JOB1)" -o sort.%J.out < sort.sh | grep -oP '\d+')

# Step 3: Stats (after sort completes successfully)
bsub -J stats -q short -W 0:30 -n 1 -w "done($JOB2)" -o stats.%J.out < stats.sh

# Step 1: Alignment
JOB1=$(sbatch --job-name=align --partition=normal --time=04:00:00 --cpus-per-task=8 --output=align.%j.out align.sh | awk '{print $4}')

# Step 2: Sort (after align completes successfully)
JOB2=$(sbatch --job-name=sort --partition=normal --time=02:00:00 --cpus-per-task=4 --dependency=afterok:$JOB1 --output=sort.%j.out sort.sh | awk '{print $4}')

# Step 3: Stats (after sort completes successfully)
sbatch --job-name=stats --partition=short --time=00:30:00 --ntasks=1 --dependency=afterok:$JOB2 --output=stats.%j.out stats.sh

Capturing the job ID

LSF: bsub prints Job <12345> is submitted to queue <normal>. — parse with grep -oP '\d+'
SLURM: sbatch prints Submitted batch job 12345 — parse with awk '{print $4}'

Interactive session¶

LSFSLURM

# Basic interactive session
bsub -Is -q interactive bash

# With resources
bsub -Is -q interactive -n 4 -M 8000 -W 2:00 bash

# Interactive with GPU
bsub -Is -q gpu -R "rusage[ngpus_physical=1]" -n 4 -M 16000 bash

# Basic interactive session
srun --pty bash

# With resources
srun --partition=interactive --cpus-per-task=4 --mem=8G --time=02:00:00 --pty bash

# Interactive with GPU
srun --partition=gpu --gpus=1 --cpus-per-task=4 --mem=16G --pty bash

Interactive jobs

In SLURM, srun --pty bash gives you an interactive shell on a compute node. You can also use salloc to allocate resources first, then srun within that allocation.