In this section we are going to discuss a program that works much like
our previous program mkrandfiles discussed in section
5.2.8. But there are a few impotant changes. First,
rather than writing separate files, we are going to open and write the
same single file, which will be accessed in parallel by all processes
of the MPI process pool. Second, instead of writing the data in small
BLOCK_SIZE integers, every process is going to lump
all the chunks into a single array and then flush it into the MPI file
in one large write.
Once we have concocted and analyzed our example program, we are going to run it and illustrate some of the elementary concepts of MPI-IO. We will use it also to test performance of the AVIDD GPFS.