Doubt regarding Cluster setupe using 3 nodes

[Gaurav Gupta]

Hello all,

I am new to MPI programming field. I am able run my first mpi program
on single machine but I have some doubt when I am running on multiple
nodes.

on single machine I tried command to run

mpirun -np 4 ./a.out
which work successfully with 4 processes in which one is master and
other 3 are workers.

Now I tried ( after enabling auto ssh login and mounting working
directory of master)
mpirun -np 4 --host 10.8.40.101 ./a.out
which runs succesfully, but my doubt is, that in this case are
resources of both local and 10.8.40.101 are utilized ? If yes than how
can I know that which process is running on which machine.

If I run
mpirun -np 4 --host 10.8.40.101,127.0.0.1 ./a.out
it this make any difference between last command and in this command.

When I tried to run above three commands, in all cases time taken by
master process is same means no performance benifit is registered. Can
anyone explain what may be reason or there is some mistake in running
the mpi application. Should I post my code here for better
explanation ?
Please guide me what can be the possible problem.

[Ilya Albrekht]

Post by Gaurav Gupta
Hello all,
I am  new to MPI programming field. I am able run my first mpi program
on single machine but I have some doubt when I am running on multiple
nodes.
on single machine I tried command to run
mpirun -np 4 ./a.out
which work successfully with 4 processes in which one is master and
other 3 are workers.
Now I tried ( after enabling auto ssh login and mounting working
directory of master)
mpirun -np 4 --host 10.8.40.101 ./a.out
which runs succesfully, but my doubt is, that in this case are
resources of both local and 10.8.40.101 are utilized ? If yes than how
can I know that which process is running on which machine.
If I run
mpirun -np 4 --host 10.8.40.101,127.0.0.1 ./a.out
it this make any difference between last command and in this command.
When I tried to run above three commands, in all cases time taken by
master process is same means no performance benifit is registered. Can
anyone explain what may be reason or there is some mistake in running
the mpi application. Should I post my code here for better
explanation ?
Please guide me what can be the possible problem.

What MPI version are you using?

You may check where processes executes by two ways:
1. Run ps -A command on both machines and see where processes executes
2. Write following shell script and execute it by mpirun command. In
output you should see host names
#!/bin/sh
hostname

PS. Don't forget that sshd should be installed and your ~ directory
should be mounted on all executing hosts.