Error with mpiexec: MPI_Comm_Size always 1!

[kralowsky]

I've followed the instructions in the 'getting started' file to set up a linux cluster with 4 nodes. Mpdboot works ok as mpdtrace verifies, and my programs compile, but don't run as they should: Taking the testc program as an example, with the command:

mpiexec -n 4 --machinefile ~/mpd.hosts ./testc

I get the output:
Hello world: rank 0 of 1 running on node1
Hello world: rank 0 of 1 running on node2
Hello world: rank 0 of 1 running on node3
Hello world: rank 0 of 1 running on node4

...so while the job is farmed out to the 4 nodes, each node thinks that it's the master, and needs to do the propagation etc. Have I missed something embarassingly obvious here? Any help would really be great!

Thanks

[grigoryzagorodnev]

Hi,

Could you provide some more details, please?

1. Exact mpdboot command

2. Output of "which mpiexec"

2. Debug output of

"mpiexec -genv I_MPI_DEBUG 3 -machinefile ~/mpd.hosts -n 4 ./testc"

Thank you in advance,

- Grigory

[kralowsky]

Hi Grigory,

Thanks for the help - it was the 'which mpiexec' that solved the problem: it was referencing /usr/bin and not the intel mpi bin directory. Pretty silly thing not to have found earlier. Thanks again.