README.md

In gongx030/bpt: A R pacakge for generating batch process templates on UMN's MSI

An R package for generating Slurm scripts

Generating Slurm script for running R commands

• In the R console: generate R and Slurm scripts: The Slurm script will request the resource of one CPU, one task, walltime of 1 hour, and run at the small queue. The script will execute at the working dir, and execute under the conda environment base.

library(bpt)
r_command <- "
print('hello world')
"
id <- 'my_script'
slurm_write_r(id, dir = '.', r_command)
slurm_write_script(
  id, 
  dir = '.', 
  nodes = 1, 
  tasks = 1, 
  partition = 'small', 
  conda = 'base', 
  wd = '.', 
  walltime = '1:00:00'
 )

The previous commands will generate my_script.r and my_script.sh under the working directory. * In the terminal: submit Slurm script.

sbatch my_script.sh

Common Slurm commands

Start a non-GPU interactive session with one node, four cores, a total memory of 32GB, walltime of 24 hours with X11 forward.

srun --nodes=1 --ntasks-per-node=4 --mem=32gb -t 24:00:00 -p interactive --x11 --pty bash

Start a GPU required interactive session with one node, one k40 GPU, walltime of 24 hours with X11 forward.

Since the k40 nodes on mesabi does not support node sharing, we may not need to explicitly specify the --ntasks-per-node and --mem (?). The website indicates that "Users are limited to a single job in the interactive and interactive-gpu partitions. However, it appears that it is OK to establish multiple interactive GPU sessions.

srun --nodes=1 -t 24:00:00 -p interactive-gpu --gres=gpu:k40:1 --x11 --pty bash

Delete a job

scancel

Show all job information

squeue -al 

Show only your job information

squeue --me

Show partition status

sinfo

gongx030/bpt documentation built on Nov. 22, 2022, 2:49 a.m.