Is there a current walkthrough for MPI Nano Clusters?

The one I found and tried tonight, by GaryExplains on YouTube didn’t work out, and I was hoping there’s an Nvidia backed version I can try out instead. This is on four nanos, which will eventually be six.

I figured I would have to make each nano have a static i.p. which I didn’t have tonight, but will try tomorrow.
examples/how_to_build_nvidia_jetson_gpu_cluster.md at master · garyexplains/examples · GitHub Is his explanation, and I keep getting stuck on ssh-copy-id function after making the private and public key successfully. It keeps popping back no identities found for the ips I found before beginning the process on all the boards. I’m also unsure how far into the path I’m supposed to use for ssh-copy-id -i /path/key.pub <ip_addr> To go around it. Thanks for any help.

I’m just getting started with mpi and nano cluster too. What I did was to set up each node with a static ip. then copied all of the node-name/ip-addresses to /etc/hosts file on each node. I used the exact same username/password on each node. Then I did the ssh-copy-id.

Then I installed the cuda-toolkit on each node. went to the var/cuda/…/samples/samplempi directory and then ran make. That put the sampleMPI program in the …/bin directory
For ease I then copied that sampleMPI file to /usr/local/bin.

Tomorrow I’ll try mpiexec and hope it has --help.

Hope you have success.

1 Like