Message-Passing Activity At a Glance.

In Figure 7-1, we see three iterations, each taking roughly 700 ms. By holding down the middle mouse button while dragging over one such iteration, you can produce the expanded view shown in Figure 7-2. More detail becomes evident. There are three important phases in each iteration, which correspond to the different axes in this three-dimensional computation. Some degree of synchronization among MPI processes is evident. Though the effect is subtle for this run, it appears that consecutive blocks of 5 processes each are synchronized at the end of phase 1 (at about 1100 ms), while every fifth process is synchronized at the end of phase 2 (at about 1300 ms). This is indicative of the benchmark running on an underlying 5x5 process grid.

Figure 7-2 Expanded View of One Iteration.