How does OpenMPI Secure SHell into all the compute

问题:

First time working with OpenMPI. I am curious how the API invokes a run-time environment to run on compute nodes. I am thinking about setting up a Linux cluster of 4 or 5 nodes. I read a lot of the documentation on creating password-less ssh access for the master node. Does OpenMPI invoke a command line argument to ssh into whatever compute nodes are declared inside the --hostfile and then begin spreading tasks?

回答1:

Open MPI does not add any additional arguments (by default) when ssh'ing to a remote machine. You should just setup your ssh keys to allow password-less logins (and if you're using an ssh passphrase, set it up so that the passphrase propagates around without manual keyboard entry).

You might also want to read these two blog entries:

http://blogs.cisco.com/performance/tree-based-launch-in-open-mpi http://blogs.cisco.com/performance/tree-based-launch-in-open-mpi-part-2