运行 正确的 Slurm 脚本有更多的节点和更少的内核
Running correctly a Slurm script with more nodes and less cores
hpc 的每个节点的最大可能内核数等于 24，但它们通常不可用，所以我想 运行 4 个节点上的代码，每个节点有 20 个内核（而不是 24)。
使用 MPI 是否正确？
#!/bin/sh
#
# Replace <ACCOUNT> with your account name before submitting.
#
#SBATCH --account=aaa # The account name for the job.
#SBATCH --job-name=job_name # The job name.
#SBATCH -N 4 # The number of nodes to use
# (note there are 24 cores per node)
#SBATCH --exclusive
#SBATCH --time=23:58:00 # The time the job will take to run.
source activate env_python
mpirun -n 80 python script.py
# End of script
这会做你想做的事：
#!/bin/sh
#
# Replace <ACCOUNT> with your account name before submitting.
#
#SBATCH --account=aaa # The account name for the job.
#SBATCH --job-name=job_name # The job name.
#SBATCH -N 4 # The number of nodes to use
# (note there are 24 cores per node)
#SBATCH --tasks-per-node=20
#SBATCH --time=23:58:00 # The time the job will take to run.
source activate env_python
mpirun -n 80 python script.py
# End of script
请求 4 个节点，每个节点有 20 个任务，将映射到 80 个 MPI 等级。 -n 80 就没有必要了。
hpc 的每个节点的最大可能内核数等于 24，但它们通常不可用，所以我想 运行 4 个节点上的代码，每个节点有 20 个内核（而不是 24)。
使用 MPI 是否正确？
#!/bin/sh
#
# Replace <ACCOUNT> with your account name before submitting.
#
#SBATCH --account=aaa # The account name for the job.
#SBATCH --job-name=job_name # The job name.
#SBATCH -N 4 # The number of nodes to use
# (note there are 24 cores per node)
#SBATCH --exclusive
#SBATCH --time=23:58:00 # The time the job will take to run.
source activate env_python
mpirun -n 80 python script.py
# End of script
这会做你想做的事：
#!/bin/sh
#
# Replace <ACCOUNT> with your account name before submitting.
#
#SBATCH --account=aaa # The account name for the job.
#SBATCH --job-name=job_name # The job name.
#SBATCH -N 4 # The number of nodes to use
# (note there are 24 cores per node)
#SBATCH --tasks-per-node=20
#SBATCH --time=23:58:00 # The time the job will take to run.
source activate env_python
mpirun -n 80 python script.py
# End of script
请求 4 个节点，每个节点有 20 个任务，将映射到 80 个 MPI 等级。 -n 80 就没有必要了。