Are the mpi_node's used automatically?

[alexisvincent]

I can't see anywhere, where the nodes are provided to mpi. Are the nodes being used automatically?

[xWuWux]

Hi,

Let's break down what's happening: In the docker-compose.yml file, there are two services defined: mpi_head and mpi_node. Both use the same openmpi Docker image. The mpi_head service has the "ports" configuration that exposes port 22, allowing you to SSH into it from your host system.

The mpi_node service doesn't have any specific port configuration. This is because it's not meant to be accessed directly but rather to be used as a worker node by the mpi_head container.

Here's how the nodes are used:

The mpi_head container, which has SSH access, is the master node. You can log into this container and use it to orchestrate and control MPI jobs.

The mpi_node containers are automatically used as worker nodes by the mpi_head container when you execute MPI jobs with mpirun. The communication between the mpi_head and mpi_node containers is handled by the OpenMPI runtime.

When you run an MPI job using mpirun from the mpi_head container, it will distribute and manage the tasks across all the nodes specified, including the mpi_node containers. You don't need to explicitly specify nodes when running the MPI job; it will be managed by OpenMPI.

So, the nodes (including mpi_node) are used automatically by the OpenMPI runtime when you run MPI applications. You interact with the mpi_head container, and it handles the distribution of tasks to the worker nodes for parallel computing.