Jeff Squyres píše v Út 15. 09. 2009 v 16:55 +0200: > On Sep 15, 2009, at 5:36 AM, Dominik Táborský wrote: > > > So I have to manually copy the compiled hello world program to all of > > the nodes so that they can be executed? I really didn't expect that... > > > > How would you expect us to execute something that doesn't exist? :-) >
I had a different idea of how it works in my mind... :-) > > So, where (in the filesystem) does the executable have to be? On the > > same place as on the master? > > > > > That's one way to do it. In general, if you specify a naked > executable name, we look for it in the PATH on each of the target > machines. We automatically append "." to your PATH on each node, so > if you "mpirun ... hello" and hello is in your CWD, then we'll find it. > > As Lenny mentioned, it's a fairly common scenario to use NFS to export > your executable to all nodes where it is needed to run. > Thanks for the answers, I highly appreciate any help! :-) Now, openmpi finds the executable, but a new problem appears - it cannot create a pipe or something. I'll try to figure it out on my own, but if I don't succeed, I'll create a new thread. Again, thank you very much, Jeff and Lenny! Dominik
