On Oct 18, 2010, at 12:08 PM, saahil.in_at_[hidden] wrote:
> > If you're just starting with MPI, we strongly recommend that you use Open MPI instead of LAM/MPI.
> Infact, I did try using OpenMPI but there were a few problems that I came across there too. I'm afraid I do not remember exactly what they were, but I stuck to LAM since then, because most of the online guides for setting up a Beowulf cluster instructed using LAM. I will try to use OpenMPI if I can, but I would like to make this work, nevertheless.
You will have much better luck getting generic support for Open MPI than LAM/MPI.
> > To directly answer your question, are you absolutely sure that your executable is named hello.o? .o files are typically intermediate object files.
> Yes, the output file was hello.o, since I compiled the C code with the -o flag as follows -
> mpicc hello.c -o hello.o
That should probably work fine, but it is "odd". Most people use .o for intermediate object files.
> Do you think that by any chance, all this is because the slave nodes and my head nodes are of different architectures, ie. 32 bit and 64 bit? Although, I made sure that the LAM version running on all of them is the same.
It is possible. Can you compile an executable on one of them and run it on all the others?
Did you compile LAM/MPI (or Open MPI) exactly the same on all nodes -- e.g., did you compile it in 32 bit mode on all machines? If not, then you are running in a heterogeneous scenario -- see the FAQ.
--
Jeff Squyres
jsquyres_at_[hidden]
For corporate legal information go to:
http://www.cisco.com/web/about/doing_business/legal/cri/
|
