intel mpi library 4.0 question

[Ian_Chivers]

I have a question regarding running mpi programs on

more than 1 system.

I have installed a 30 day evaluation copy of

release 4 of the library on two

systems. They both run

Windows Vista Home Premium 64 bit.

The following shows that I can actually run

the sample intel program on these two systems.

C:\\document\\fortran\\newbook\\mpi\\examples>mpiexec -hosts 1 hornbeam 4 intel_test

Hello world: rank 0 of 4 running on hornbeam

Hello world: rank 1 of 4 running on hornbeam

Hello world: rank 2 of 4 running on hornbeam

Hello world: rank 3 of 4 running on hornbeam

C:\\document\\fortran\\newbook\\mpi\\examples>mpiexec -hosts 1 dell-xps 4 intel_test

Hello world: rank 0 of 4 running on dell-xps

Hello world: rank 1 of 4 running on dell-xps

Hello world: rank 2 of 4 running on dell-xps

Hello world: rank 3 of 4 running on dell-xps

C:\\document\\fortran\\newbook\\mpi\\examples>

But when I try to run on BOTH systems I get

C:\\document\\fortran\\newbook\\mpi\\examples>mpiexec -hosts 2 dell-xps 2 hornbeam 2

intel_test

Hello world: rank 0 of 4 running on dell-xps

Hello world: rank 1 of 4 running on dell-xps

and the program hangs.

I can't find anything in the documentation to give a clue a to

what might be the problem here.

Any suggestions welcome.

Ian Chivers

[Gergana_S_Intel]

Hi Ian,

Have you taken a look at the Getting Started Guide for the Intel MPI Library? It contains troubleshooting techniques to help you figure out what might be happening.

Since you seem to be ok running on a single node, I'd say check to make sure you have registered your credentials on the cluster. You can do that via the wmpiregister GUI utility, or simply running mpiexec -register.

Also, check to make sure the smpd service is running on all nodes of the cluster. You can do so via smpd -hosts.

Let us know how it goes.

Regards,
~Gergana

[Ian_Chivers]

I have run the wmpiregister utility as you mentioned.
The output is the same as in the original post.

when running

smpd -hosts

I only get back the local host.

I have read the Getting Started Guide.

Cheers

Ian

[Ian_Chivers]

I forgot to mention that I did a

[CTRL] [ALT] [DELETE]

on the other system (hornbeam)
and when looking at the task manager the smpd
process is running.

Sorry

Ian

[Dmitry_K_Intel2]

Hi Ian,

Could you try to run:
smpd -sethosts dell-xps hornbeam
Check for available hosts:
smpd -hosts
2 hosts should be listed. If you don't see both hosts in this list the potential problem either firewall or anti-virus (I hope these 2 machines are connected via ehternet)

Regards!
Dmitry

[Ian_Chivers]

Thanks for getting back.
The firewall is OFF on both machines.

Here is the output from what you suggested.

C:\document>smpd -sethosts dell-xps hornbeam

[-1:5540]....ERROR:Unable to open the HKEY_LOCAL_MACHINE\SOFTWARE\Intel\MPI\SMPD

registry key, error 5

Access is denied.

Error: unable to save the hosts data.

C:\document>

Cheers

Ian

[Dmitry_K_Intel2]

For those who meet the same problem:

Run your application on different nodes with I_MPI_DEBUG=9 and compare the output.
In Ian's output I see:
PROCESSOR_IDENTIFIER=Intel64 Family 6 Model 26 Stepping 4, GenuineIntel
for one node and:
PROCESSOR_IDENTIFIER=Intel64 Family 6 Model 15 Stepping 7, GenuineIntel
for another one and different settings for collective operations.

To avoid this set I_MPI_PLATFORM env variable to 'auto' - something like:
mpiexec -genv I_MPI_PLATFORM auto ...

Regards!
Dmitry