In lectures encouraging people to use the new MPI-2 Standard for parallel I/O (see Chapter 7 of ), we find it useful to compare independent I/O with collective I/O on the same problem of writing a (block,block)-distributed array to disk. The many small writes done by the independent writes, which mimic the Unix interface, are illustrated in Figure 6 . The many individual writes can be seen (the white sections are just the borders of small writes close together; they can be resolved by zooming in further). The same problem carried out with the MPI-2 collective I/O operations show how small communications and computational overheads can be traded for large amounts of I/O overhead (see Figure 7 . Figure 6 is a zoomed-in view; the elapsed time for the entire I/O operation is about 27 seconds. In Figure 7 the elapsed time can be seen to be about 2 seconds, and more than half of this time can be seen to be taken up by the open operation. These runs were done on an Intel Paragon.