Introduction
The GRID infrastructure offers a great deal of cheap computing resources, which are not very uniform though in terms of underlying software stack. This is especially true when a code needs to run in parallel via MPI. A workaround we developed in order to run different codes across shared memory (mainly SMP) resources on the GRID is the so-called "relocatable package" approach.
With this approach both a precompiled OpenMPI distribution and the code binaries compiled with it are sent to the Worker Node with every job and deployed in order to work properly. From the GRID point of view, the user needs just to submit a standard (non-MPI) job in the usual way.
The precompiled packages should be able to run in any reasonably recent environment and the big advantage to the user is that he can completely ignore the actual software stack on Worker Node. This approach works well on SMP resources, but has no support for distributed memory computers connected via high-speed networks (t.i. clusters).
Here below are some How-Tos that can help you to run some of the applications we already packaged and used with the "relocatable approach".