Sorry for entering the discussion late. From tracing the email thread, I
somewhat gather the following:
1. you have installed Open MPI 1.1.2 on two 686 boxes
2. you created a hostfile on one of the nodes and execute mpirun from that
node. You gave us a prefix indicating where we should find the Open MPI
executables on each node
3. you were getting an error message indicating that mpirun was unable to
find your executable
4. you didn't encounter this problem when running on a cluster
If I have those facts correct, then the problem is simple. My guess is that
the cluster you were using has a shared file system - hence, the remote
nodes "see" your executable in the same relative location across the
In your simple setup with the two boxes, it sounds like you don't have a
shared file system. When mpirun attempts to locate the executable on
bernie-3, it won't find the file since it doesn't exist on that node's file
system. Once you copied the file over to bernie-3, then mpirun could find it
so everything works fine.
We hope to add file pre-positioning at some point in the future for systems
such as yours. However, that is some time away due to priorities. For now,
Open MPI requires that your executable (and the Open MPI executables and
libraries) be available on each node you are trying to use.
Hope that helps to explain the problem.
On 1/2/07 2:03 PM, "jcolmenares_at_[hidden]" <jcolmenares_at_[hidden]> wrote:
> I had configured the hostfile located at
> I copied the file to bernie-3, and it worked...
> Now, at the cluster I was working at the Universidad de Los Andes
> (Venezuela) -I decided to install mpi on three machines I was able to put
> together as a personal proyect- all I had to do was to compile and run my
> applications, that is, I never copied any file to any other machine...
> now, I had to. I'm sorry if it was obvious and made you guys loose some
> time, but why on a cluster I didn't have to copy any files, and now I must
> do so?
> Thanks for you patiance!
> users mailing list