0

I am testing out OpenMPI, provided and compiled by another user, (I am using soft link to his directories for all bin, include, etc - all the mandatory directories) but I ran into this weird thing:

First of all, if I ran mpirun with -n setting <= 10, I can run this below. testrunmpi.py simply prints out "run." from each core.

# I am in serverA.
bash-3.2$ /home/karl/bin/mpirun -n 10 ./testrunmpi.py
run. 
run. 
run. 
run. 
run. 
run. 
run. 
run. 
run. 
run.

However, when I tried running -n more than 10, I will run into this:

bash-3.2$ /home/karl/bin/mpirun -n 24 ./testrunmpi.py
karl@serverB's password: Could not chdir to home directory /home/karl: No such file or directory
bash: /home/karl/bin/orted: No such file or directory
--------------------------------------------------------------------------
A daemon (pid 19203) died unexpectedly with status 127 while attempting
to launch so we are aborting.

There may be more information reported by the environment (see above).

This may be because the daemon was unable to find all the needed shared
libraries on the remote node. You may set your LD_LIBRARY_PATH to have the
location of the shared libraries on the remote nodes and this will
automatically be forwarded to the remote nodes.
--------------------------------------------------------------------------
--------------------------------------------------------------------------
mpirun noticed that the job aborted, but has no info as to the process
that caused that situation.
--------------------------------------------------------------------------
bash-3.2$
bash-3.2$
Permission denied, please try again.
karl@serverB's password:
Permission denied, please try again.
karl@serverB's password:

I see that the work is dispatched to serverB, while I was on serverA. I don't have any account on serverB. But if I invoke mpirun -n <= 10, the work will be on serverA.

This is strange, so I checked out /home/karl/etc/openmpi-default-hostfile, and tried set the following:

serverA slots=24 max_slots=24
serverB slots=0 max_slots=32

But the problem persists and still gives out the same error message above. What must I do in order to have my program run on serverA only?

Karl
  • 5,613
  • 13
  • 73
  • 107

1 Answers1

1

The default hostfile in Open MPI is system-wide, i.e. its location is determined while the library is being built and installed and there is no user-specific version of it. The actual location can be obtained by running the ompi_info command like this:

$ ompi_info --param orte orte | grep orte_default_hostfile
MCA orte: parameter "orte_default_hostfile" (current value: <LOOK HERE>, data source: default value)

You can override the list of hosts in several different ways. First, you can provide your own hostfile via the -hostfile option to mpirun. If so, you don't have to put hosts with zero slots inside it - simply omit machines that you have no access to. For example:

localhost slots=10 max_slots=10
serverA slots=24 max_slots=24

You can also change the path to the default hostfile by setting the orte_default_hostfile MCA parameter:

$ mpirun --mca orte_default_hostfile /path/to/your/hostfile -n 10 executable

Instead of passing each time the --mca option, you can set the value in an exported environment variable called OMPI_MCA_orte_default_hostfile. This could be set in your shell's dot-rc file, e.g. in .bashrc if using Bash.

You can also specify the list of nodes directly via the -H (or -host) option.

Hristo Iliev
  • 72,659
  • 12
  • 135
  • 186