> There is nothing preventing you from creating your own shared memory
> for use by your application. This does not impact the MPI implementation.
> You need to make sure you take care of all the issues associated with
> multiple procs/threads accessing the same cache lines, which can cause
> a bit of performance degradation, if not managed.
How can we create "our own shared memory segment?" Do you mean Linux System
calls? I see shmget routine, for example, but it is a C routine. Do we have
to write a fortran interface to use it?