There is a special wrapper-compiler which takes care of specifying the exact place for all the necessary libraries. Compile your program with

$ hcc myprogram.c -o myprogram -lmpi

First you need to decide whether you are going to use more than one physical computer for your program. If you are just testing your MPI application for errors during development phase, you are encouraged to use one physical processor with many virtual processors (just regular processes).

If so, boot your (virtual) supercomputer with

$ lamlocal

and you will be allocated 20 virtual processors on the machine you ran the command.

If you need many processors to cooperate (as a cluster), you will need to specify a file containing all machines which are to participate. You need to specify a file containing all machines which are to participate. Use /global/etc/banachcluster for example. Start with logging into one of the machines contained in this file and follow the commands below. Please remember that if you run computations which are intensive in communication, you will slow down the network performance student lab considerably (affecting all other regular users of the lab)! Use lamlocal if you don't need a real cluster.

$ export LAMRSH=“ssh -x” $ nice lamboot /global/etc/banachcluster

The file banachcluster contains 26 machines. You may also use nullcluster, which is also 26 machines, and also mathcluster which is banachcluster and nullcluster concatenated (52 machines).

Each user must boot the cluster for himself. Your booting will not affect any other users running computations (though you must share memory and processor time).

If no errors occured, the cluster is ready. Use lamnodes to check the list of computers in your booted cluster.

There is a special wrapper-compiler which takes care of specifying the exact place for all the necessary libraries. Compile your program with

$ hcc myprogram.c -o myprogram -lmpi

Run your program on for example 4 processors with the command

$ mpirun -np 4 myprogram

When you are finished, please shut down your cluster using lamhalt.