I am trying to use MPI in a cluster and would like to be able to control which ranks get scheduled in which nodes.
Note: I am using OpenMPI 2.1.0.
For that I am using a rankfile. If I use the following rankfile:
ubuntu@ip-172-31-8-16:~/dist_log_reg$ cat rankfile
rank 0=localhost slots=1
rank 1=54.153.103.12 slots=1
I get:
ubuntu@ip-172-31-8-16:~/dist_log_reg$ mpirun -v -np 1 -rankfile rankfile hostname
--------------------------------------------------------------------------
The rankfile that was used claimed that a host was either not
allocated or oversubscribed its slots. Please review your rank-slot
assignments and your host allocation to ensure a proper match. Also,
some systems may require using full hostnames, such as
"host1.example.com" (instead of just plain "host1").
Host: ip-172-31-8-16
If I use only one entry in the rankfile:
ubuntu@ip-172-31-8-16:~/dist_log_reg$ cat rankfile
rank 0=localhost slots=1
I get:
ubuntu@ip-172-31-8-16:~/dist_log_reg$ mpirun -v -np 1 -rankfile rankfile hostname
--------------------------------------------------------------------------
All nodes which are allocated for this job are already filled.
I have tried everything I can think of (e.g., installing other distributions of MPI and trying different options in the rankfile) but haven't been able to make this work.
Any ideas?
I managed to create your error with passing localhost
as hostname
. But When I use the actual system-name I managed to run it.
rank X=myPC slot=Y
I believe that Open MPI probes the hostnames and performs a gethostname call.