Super User is a question and answer site for computer enthusiasts and power users. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I'm trying to create really simple "cluster" from 2 multicore computers using openmpi.

My problem is that I can't find any tutorials on that matter. I don't want to use torque because it's not necessary in my case nevertheless all tutorials give configuration details either about torque or mpd (which doesn't exist in openmpi implementation).

Could you give me some tips or links to appropriate manuals?

Steps I've already completed:
- openmpi installation
- network configuration (computers see each other)
- ssh password-less login to second computer

I tried using machinefiles without further configuration and with just 2 IPs in it. But jobs don't seem to start at all after initialization part. (MPI seems to work because I'm able to scatter jobs on multiple cores of both computers without communication between them).

share|improve this question
Do you get any error messages from OpenMPI? – chuck Oct 16 '12 at 13:08

The best solution to use OpenMPI on such a small number of machines is to use the mpirun or mpiexec commands, with or without a host file. This should work and you don't require a third party process manager such as Torque or Slurm.

If your MPI job is not starting, there might be something wrong with the communication. Test this by creating a host file that only contains your local machine, which should assign all your MPI processes to your localhost. Also, before running your own MPI job, start with the example jobs provided with Open MPI.

It can be quite finicky to configure Open MPI. In your configure script, expressly state which underlying communication support you require (for example sm and tcp). When you're running your job, specify which channels MPI should use. By explicitly specifying these values, MPI will return an error if there is something wrong.

To get started with mpirun, have a look at the Open MPI documentation of the latest version:

share|improve this answer

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .