Some MPI guidelines

In order to run the parallel version of PUMA you need access to an MPI Grid. MPI Grids are very common, but if you don't know any, you can setup one using your own computers. This webpage will give you a few guidelines for setting up your own MPI Grid.

The instructions given here are directed to Linux users, inspite that, Windows and MacOS implementations are very similar to configure and use.

Starting up an MPI Grid

Consider the following scenario:

1. Two computers with IPs 192.168.0.1 and 192.168.0.2 respectively, with two processors each and running a Linux distribution.
2. An installed version of the LAM/MPI implementation.

We want to setup a grid with four nodes, each node corresponding to one avaliable processor.

To startup a grid do the following:

1. Create a file named boot with the following lines:
   

   192.168.0.1
   192.168.0.1
   192.168.0.2
   192.168.0.2
   

2. Execute:

   lamboot boot

3. Now you have an MPI Grid with four nodes. Two nodes are on IP 192.168.0.1 and two other nodes are on IP 192.168.0.2

Submitting an application to the grid

Be sure you compiled your MPI application with mpicc (or mpic++, or mpif77, depending on in which language your application was written).

Simply type:

mpirun C <application> <application arguments>

where the C stands for "using all the available processors", <application> represents the name of the application you are trying to run (in this case puma-par), and <application arguments> represents the parameters of puma-par.

---