Installation of the MPI version

[Aloïs Glibert]

Good morning,

I'm trying to install the MPI version of nemo2.3.54 on some ubuntu (20.10 or 19.04) machines, but cannot seem to get it to work. The install seems to go just fine, no terrible warning and the executable is produced, but when I launch any simulation given in example, I get

N E M O 2.3.54 [18 Jan 2021]

  Copyright (C) 2006-2020 Frederic Guillaume
  This is free software; see the source for copying
  conditions. There is NO warranty; not even for
  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
  http://nemo2.sourceforge.net
------------------------------------------------
reading parameters from "../example/multiselection.ini"
setting random seed from input value: 234

--- SIMULATION 1/1 ---- [ delet-quanti-selection ]

    start: 02-02-2021 10:32:35
    mode: overwrite
    traits: delet, quant
    LCEs: breed(1), viability_selection(2), save_stats(3), save_files(4), aging(5),
    outputs: test/multiselection/{*.log, *.txt}

--------------------------------------------------------------------------
MPI_ABORT was invoked on rank 0 in communicator MPI_COMM_WORLD
with errorcode 101.

NOTE: invoking MPI_ABORT causes Open MPI to kill all MPI processes.
You may or may not see output from other processes, depending on
exactly when Open MPI kills them.
--------------------------------------------------------------------------

Any idea how I could make it work ?

Thank you for your time

Aloïs

[Frederic Guillaume]

Hi Aloïs

That abort may be caused by too few or too many CPUs allocated to your job. The MPI version should be allocated exactly n + 1 CPUs, where n is the number of replicates in your simulation, or a divider of it. This is because one CPU is reserved for the "manager", and n are for the "workers" who perform the simulations. The manager only collects the results from the workers and assign the replicates to them.

For instance, if you want to run nemo2.3.54mpi for a simulation wit 50 replicates, ask for 26, or 51 CPUs (mpirun -n26 nemo2.3.54mpi ... ) then nemo will divide the number of replicates among the 25 or 50 worker CPUs.

Please report the command you used to start the MPI job if this doesn't work. Also, make sure that the CPUs are correctly assigned to your MPI job.

Best,

Fred

[Aloïs Glibert]

Hi,

I totally missed the fact that I had to use mpirun to run it, which was my problem :D From then on, no more problems, everything works fine :D Have a great day :)

Aloïs