Raymond Wan wrote: > > Hi Jeff, > > Some "good" news (but still some bad news). Y and Z are part of a set > of 8 machines and I found out that mpirun works for one of them. I > didn't checked a couple of them before -- sorry! However, I'm no closer > to the solution since all 8 should be "identical", according to our > sysadmin. He said the only difference (that he can think of) between > the working one and all the others is that the working one has an NIS > server installed. It is the NIS server for the cluster (presumably, the > others run a client version). Could that be the reason? He can't think > of anything else that distinguishes between them but he says it is > possible that the NIS server is correctly configured for what we use it > for, but not for what I'm doing with Open MPI -- he doesn't know what > should be done, though.
In an earlier e-mail in this thread, I theorized that this might be a problem with your name service. This latest information seems to support that theory. To test, on all 3 systems, use the 'host' command to see if you can resolve the hostnames of all the 3 systems. On host X, do this: host X host Y host Z Then do the same on hosts Y and Z. If the 'host' command can resolve properly, you should see something like this: $ host foo foo.example.com has address 192.168.1.1 If 'host' can't resolve a hostname properly, you should see something like this: $ host bar Host bar not found: 3(NXDOMAIN) OpenMPI should be using the same nameservice libraries all the other programs use, so I find it hard to believe everything *but* OpenMPI is working propery, but I suppose it could be possible. I've seen weirder. -- Prentice
