kenji-h/gcp-mpi-cluster

MPI cluster on GCP

gcp-mpi-cluster's Issues

make -np option configurable

You can set the number of processes when running MPI programs.

mpirun -np 10 ./main

Even when running on cluster, it would be nice if you can configure the number of processes per node, which defaults to 1.

Remove redundant become directives

There are many become in ansible scripts.
Setting become on role looks simper.

You can set the become options per:
　・playbook
　・role
　・task

https://stackoverflow.com/questions/37200337/is-there-anyway-to-run-multiple-ansible-playbooks-as-multiple-users-more-efficie

pass program argument when compiling

It's good if you can specify source code to compile.

• add app directory in top directory
• user can add source code in app directory
• user can remotely compile source code in app directory

yum loop deprecated

change to use array and remove the loop.

[DEPRECATION WARNING]: Invoking "yum" only once while using a loop via squash_actions is deprecated. Instead of using a loop
to supply multiple items and specifying `name: {{ item }}`, please use `name: ['emacs', 'mpich', 'mpich-devel']` and remove
the loop. This feature will be removed in version 2.11. Deprecation warnings can be disabled by setting
deprecation_warnings=False in ansible.cfg.

network configuration

create firewall rules:

• allow ssh from only specified IPs
• allow all communications inside of the network

make the number of nodes configurable

For now, it only construct a cluster of a small size . But you will want to try larger size of cluster soon.
It is better to have the number of nodes configurable.

use cpp instead of c

I found /usr/lib64/mpich/bin/mpic++.
Using cpp would be better than using c.

Cannot send/receive data in cluster

Master can get its own rank and processor_name .
But it stalls when sending/receiving data from slaves.

add MPI commands to PATH

Sometimes you log in a node and try command by yourself.
For this purpose, adding /usr/lib64/mpich/bin/ to PATH is a good idea.

UNREACHABLE before server launch

create_cluster runs ansible scripts after terraform.
If the server is not created yet, when executing ansible scripts, It throws errors.