Program: launch
Developer: Lisa L. Lowe, [email protected]
Originally for OIT-HPC@NCSU
Purpose: Bundles a series of serial jobs by distributing the commands listed in a file
To use on Henry2 via LSF (this goes in your batch script):
mpirun ./launch [input file]
For general use:
mpirun -n [MPI tasks] ./launch [input file]
Get the application and enter the directory:
git clone https://github.com/lisalenorelowe/ser2launch.git
cd ser2launch
Compile the application, this assumes an Intel MPI environment:
module load PrgEnv-intel
source makeit.sh
For extra write statements to check the filename and that commands were sent/recieved properly, change makeit.sh to compile the application with "debug".
#For extra debugging, uncomment this line in makeit.sh
mpiif90 -o launch -Ddebug launch.F90
Run the application:
First run the test case. The example commands are in commands.txt. The commands say Hello from the node they are running on, write the date, and then sleep for 3 seconds.
To submit as a job on Henry2:
bsub < launch.csh
You should see this type of output in the LSF output launch.out.[JOBID]
Sending task 76 of 203
Hello from n2e2-8 on Sat Aug 1 10:00:44 EDT 2020
To run your own cases, either modify commands.txt, or change the name of the input file. For example, if your list of commands is in a file my_commands.txt, then change the line in launch.csh to:
mpirun ./launch my_commands.txt
In the Fortran code, the number of characters allowed in a command line is 200. You can modify the code directly to change that number. If you do that, remember to recompile the code.
Acceptable Use
Check your local HPC documentation for instructions on loading an MPI environment and submitting jobs. Also, if all of your tasks take 5 minutes and one takes an hour, many cores will be idle while waiting on the outstanding task to finish. That is usually not considered as acceptable use. Please be mindful of this.
For short tasks with overhead
If you are running a very short task using a software that takes a while to load, it can cause problems or even crash a node. In particular, I've seen this with MATLAB. If you have a very simple and short task, try built in loop parallelization for the software, e.g., foreach/dopar(R) and search up keywords 'parallel pool' for the latest Python and MATLAB recommendations.
This repo includes a sample MATLAB parallel loop: par_matlab.
Also
CONSIDER THIS A BETA VERSION! There is no error checking of individual commands as of yet. Use of this software is at your own risk.