728x90
반응형
기타 SBATCH 옵션
- Dependency Job 수행 방법
#SBATCH --error=파일이름
#SBATCH --mail-type=ALL
#SBATCH --mail-user=이메일 주소
#SBATCH --cpu-per-task=쓰레드 수
#SBATCH --mem=메모리 크기 // --mem, --mem-per-cpu는 중복 불가, 노드 설정에 RealMemory=8080(M) 부분 반영 필요
#SBATCH --mem-per-cpu=메모리 크기
#SBATCH --gres=gpu:개수
#SBATCH --time=HH:MM:SS
#SBATCH --begin=시작시간
#SBATCH --partition=파티션이름
#SBATCH --qos=QOS 이름
#SBATCH --nodelist=노드 리스트
#SBATCH --exclude=노드 리스트
#SBATCH --get-user-env
#SBATCH --export=ALL
#SBATCH --test-only
#SBATCH --array=0-9
#SBATCH --priority=우선순위
#SBATCH --requeue
#SBATCH --exclusive ##독점 모드
#SBATCH --account
#SBATCH --dependency=[after|afterany|afterok|afternotok:<jobid>] 이전작업에 종속성을 설정
예시 작업(1)
$ cat dep_job_01.sbatch
#!/bin/bash
#SBATCH --job-name=dep_job_01
#SBATCH --ntasks=1
#SBATCH --output=dep_job_01.out
#SBATCH --partition=cpu.q
#SBATCH --get-user-env
# Run your command
sleep 60
$ sbatch dep_job_01.sbatch
$ squeue
JOBID PARTITION NAME USER ST TIME NODES NODELIST(REASON)
12 cpu.q dep_job_ estar987 R 0:04 1 slurm91
예시 작업(2)
- 파티션 두 개 동시 설정
$ cat dep_job_011.sbatch
#!/bin/bash
#SBATCH --job-name=dep_job_01
#SBATCH --ntasks=1
#SBATCH --output=dep_job_01.out
#SBATCH --partition=cpu.q,docker.q
#SBATCH --get-user-env
# Run your command
sleep 60
$ sbatch dep_job_01.sbatch
$ squeue
JOBID PARTITION NAME USER ST TIME NODES NODELIST(REASON)
13 cpu.q dep_job_ estar987 R 0:03 1 slurm91
예시 작업(3)
- 이전 작업에 종속성을 설정
$ sbatch --dependency=afterok:2016 dep_job_011.sbatch
Submitted batch job 14
$ sbatch --dependency=afterok:2017 dep_job_011.sbatch
Submitted batch job 15
$ sbatch --dependency=afterok:2018 dep_job_011.sbatch
Submitted batch job 16
$ sbatch --dependency=afterok:2019 dep_job_011.sbatch
Submitted batch job 17
$ sbatch --dependency=afterok:2020 dep_job_011.sbatch
Submitted batch job 18
$ squeue
JOBID PARTITION NAME USER ST TIME NODES NODELIST(REASON)
14 cpu.q dep_job_ estar987 R 0:18 1 slurm91
15 cpu.q dep_job_ estar987 R 0:12 1 slurm92
16 cpu.q dep_job_ estar987 R 0:09 1 slurm91
17 cpu.q dep_job_ estar987 R 0:06 1 slurm92
18 cpu.q dep_job_ estar987 R 0:03 1 slurm91
예시 작업(4)
- Array Job 수행 방법
$ cat array_job.01.sbatch
#!/bin/bash
#SBATCH --job-name=array_job_01
#SBATCH --ntasks=1
#SBATCH --output=array_job_01_%A_%a.out
#SBATCH --partition=docker.q
#SBATCH --get-user-env
#SBATCH --array=1-10
#SBATCH --account=prj_estar987
# Run your command
sleep 5
echo "Array Task ID: $SLURM_ARRAY_TASK_ID"
$ sbatch array_job.01.sbatch
$ squeue
JOBID PARTITION NAME USER ST TIME NODES NODELIST(REASON)
29_1 docker.q array_jo estar987 R 0:12 1 docker95
29_2 docker.q array_jo estar987 R 0:12 1 docker96
29_3 docker.q array_jo estar987 R 0:12 1 docker95
29_4 docker.q array_jo estar987 R 0:12 1 docker96
29_5 docker.q array_jo estar987 R 0:12 1 docker95
29_6 docker.q array_jo estar987 R 0:12 1 docker96
29_7 docker.q array_jo estar987 R 0:12 1 docker95
29_8 docker.q array_jo estar987 R 0:12 1 docker96
29_9 docker.q array_jo estar987 R 0:12 1 docker95
$ ls -1 array_job_01*.out
array_job_01_29_10.out
array_job_01_29_1.out
array_job_01_29_2.out
array_job_01_29_3.out
array_job_01_29_4.out
array_job_01_29_5.out
array_job_01_29_6.out
array_job_01_29_7.out
array_job_01_29_8.out
array_job_01_29_9.out
array_job_01_40_10.out
array_job_01_40_1.out
array_job_01_40_2.out
array_job_01_40_3.out
array_job_01_40_4.out
array_job_01_40_5.out
array_job_01_40_6.out
array_job_01_40_7.out
array_job_01_40_8.out
array_job_01_40_9.out
$ cat array_job_01_40_4.out
Array Task ID: 4
- array 구문 활용 예시
#!/bin/bash
#SBATCH --job-name=array_job_01
#SBATCH --ntasks=1
#SBATCH --output=array_job_01_%A_%a.out
#SBATCH --partition=docker.q
#SBATCH --get-user-env
#SBATCH --array=1,3,5,7,9 // 수정
#SBATCH --account=prj_estar987
# Run your command
sleep 5
echo "Array Task ID: $SLURM_ARRAY_TASK_ID"
$ squeue
JOBID PARTITION NAME USER ST TIME NODES NODELIST(REASON)
50_1 docker.q array_jo estar987 R 0:03 1 docker95
50_3 docker.q array_jo estar987 R 0:03 1 docker96
50_5 docker.q array_jo estar987 R 0:03 1 docker95
50_7 docker.q array_jo estar987 R 0:03 1 docker96
50_9 docker.q array_jo estar987 R 0:03 1 docker95
#!/bin/bash
#SBATCH --job-name=array_job_01
#SBATCH --ntasks=1
#SBATCH --output=array_job_01_%A_%a.out
#SBATCH --partition=docker.q
#SBATCH --get-user-env
#SBATCH --array=1-20:2
#SBATCH --account=prj_estar987
# Run your command
sleep 5
echo "Array Task ID: $SLURM_ARRAY_TASK_ID"
$ squeue
JOBID PARTITION NAME USER ST TIME NODES NODELIST(REASON)
65_1 docker.q array_jo estar987 R 0:02 1 docker95
65_3 docker.q array_jo estar987 R 0:02 1 docker96
65_5 docker.q array_jo estar987 R 0:02 1 docker95
65_7 docker.q array_jo estar987 R 0:02 1 docker96
65_9 docker.q array_jo estar987 R 0:02 1 docker95
65_11 docker.q array_jo estar987 R 0:02 1 docker96
65_13 docker.q array_jo estar987 R 0:02 1 docker95
65_15 docker.q array_jo estar987 R 0:02 1 docker96
65_17 docker.q array_jo estar987 R 0:02 1 docker95
65_19 docker.q array_jo estar987 R 0:02 1 docker96
반응형
'HPC' 카테고리의 다른 글
Singularity (0) | 2024.11.25 |
---|---|
[Linux] ANSYS 소프트웨어 설치 (0) | 2024.11.18 |
[SLURM] SBATCH(1) (0) | 2024.09.06 |
[SLURM] SRUN 명령어 (1) | 2024.09.05 |
[Ubuntu] SGE 설치 (0) | 2024.09.03 |