using MPI/LAM and trying to run example1

[gortiz]

Dear Scalapack forum,

Recently I have compiled tests program included into scalapack_installer
distribution. And also in order to run scalapack programs using MPI/LAM
I have used lamboot command: "lamboot -v /usr/local/etc/lam-bhost.def"
with file "lam-bhost.def" containing node definitions.

Then, I have compiled example1.f downloaded from netlib webpage doing:
mpif77 -o example1 example1.f ~/Scalapack/scalapack_installer_0.91/lib/libscalap
ack.a ~/Scalapack/scalapack_installer_0.91/lib/blacsF77.a ~/Scalapack/scalapack_installer_0.91/lib/blacs.a ~/Scalapack/scalapack_installer_0.91/lib/blacsF77.a ~/Scalapack/scalapack_installer_0.91/lib/libreflapack.a /usr/lib64/libblas.a /usr/local/lib/libmpi.a

Up to here all seems OK, But, when I trying to run this example1, the following output I obtain:

mpirun -v -np 6 example1
26200 example1 running on n0 (o)
mpirun: cannot start example1 on n1: No such file or directory

And also doing a copy of executable example1 in all nodes the output
is the same...

Somebody can help me?

Thanks a lot
g.

[Danesh_D]

Gortiz,

I think you have posted your problem into wrong forum. You can get better feedback/solution if you put your post in MPI/LAM forum. Anyway, I write my suggestions. Hope it helps. The first thing that appears to be wrong is that you should write:

mpirun -np 6 ./example1

unless you have put "example1" in path. If you tried this solution and still have problems keep reading. So, it seems that there is something wrong with your MPI installation. Do as following:

1. First test your program on your local machine and forget about other nodes by running "mpirun -np 2 ./example1" and see if it runs.

2. Try MPI Hello World program which is very simple and used to run on any system. Let see if you can run it and they jump into your own samples.

3. You should define your hosts within a host file and then launch "mpirun" like this:

mpurin -np 4 --hostfile my_hosts ./prog

without this, you will just run your code on your local machine with several threads which run concurrently and not in parallel. If you need to define your hosts in your "hostfile" see this link http://www.open-mpi.org/faq/?category=r ... e-spmd-run

Danesh

[gortiz]

Thanks a lot Danesh!

Yes it is true about Forum matter you said. But
it is also very nice that people like you try to help anyway!

Then, let me say that I went to proper Forum and found
that MPI/LAM seems to be a little different with some
funtionallity. For example, it is possible tell to mpirun
where is located a copy of program that you want to run, i.e.

mpirun -s n0 -n6 ./example1

will take a copy of executable "example1" from node0 and distribute
they to six of nodes listed in host_file.

Best regards,

g.

[Danesh_D]

Gortiz,

If you still have problem using LAM/MPI, I would suggest you to move to Open-MPI since Open-MPI is the best MPI implementation which has combined other technologies from other libraries including LAM/MPI. Besides, you may have full implementation of MPI-2 while LAM/MPI is not full compatible with MPI-2. Using Open-MPI is very simple:

1- Download proper file from their web site.
2. Install it.
3. Test if "ompi_info" does work, so installation was successful.
4. setup all your nodes so they can login without being prompted for password via SSH to each other.
5. Setup a host file where you define master and slave nodes with their IPs and slots.
6. Disable firewall on all your nodes (or set the firewall to allow all traffic from other nodes since Open-MPI choose a random port each time it runs)
7. Run your program using "mpirun".

Open-MPI is a daemon based program and 100% cluster friendly so you will not loose anything.

Good Luck,

Danesh