First of all STOP USING LLNL WEB PAGES. They are two years old. You want srun's multi-prog option. See: http://www.schedmd.com/slurmdocs/srun.html
There are several tutorials here: http://www.schedmd.com/slurmdocs/tutorials.html Quoting [email protected]: > > Dear SLURM development team, > > Could you please direct me to a tutorial designed for researchers who are > not already proficient in > shell programming. > > My expertise is physics. I learned FORTRAN in order to program the > computations that I need to > perform. But now I am trying to run my FORTRAN programs on our > supercomputers (CCNI) and need to learn how to use SLURM. > > I've read through your website: > https://computing.llnl.gov/linux/slurm/help.html > > And I've watched the tutorials on Youtube. > > But they are not accessible to me, and I fear that it'll take another > year devoted to learning BASH, SLURM, etc., before I am able to run my > FORTRAN programs (which took a year for me to program). > > All that I want to do is request N nodes, and run a different executable > on each node. > > I wish that the commands were: > > GIVE ME N nodes > DO i=1,N > RUN program_i on node i > END DO > END > > But all of the examples that I've seen show requesting multiple nodes to > run one program. > > My question seems most similar to SLURM FAQ #8: > "How can I run multiple jobs from within a single script?" > But to me, the answer seems to say, "You can. People do." > I can't seem to find the part that says how to do it. > > The YouTube tutorials seem designed for system administrators. > > I'm sure the problem is that I am just a complete beginner at this stuff. > If I was fluent in BASH, I > imagine that the examples online would make sense. But I am hoping that > it is possible to use > SLURM without first becoming fluent in BASH. > > I'm hoping there is a tutorial for researchers who are not fluent in > BASH, preferably with example > scripts that are not overly complicated with dozens of flags and dozens > of environment variables. > > If not, then after I figure it out, I would be happy to contribute one > for other researchers like me. > But I'm hoping that someone has already done that. > > > Thank you, > Joseph > > >
