So far, we've talked about getting PE working, creating message passing parallel programs, debugging problems, and debugging parallel applications. When we get a parallel program running so that it gives us the correct answer, we're done. Right? Not necessarily. In this area, parallel programs are just like sequential programs; just because they give you the correct answer doesn't mean they're doing it in the most efficient manner. For a program that's relatively short running or is run infrequently, it may not matter how efficient it is. But for a program that consumes a significant portion of your system resources, you need to make the best use of those resources by tuning its performance.
In general, performance tuning can be a complicated task. The Hitchhiker's Guide to the Galaxy tells us that dolphins were more intelligent than humans. When the Earth was being destroyed to make way for a hyperspace bypass, the dolphins were able to escape the planet and made one more attempt to communicate. Their parting words to humans were again misinterpreted as a water trick although the actual message was "So long and thanks for all the fish". The people that tune the performance of applications regard the original developers of their code in much the same way (even if they, themselves, were the original developers). They appreciate the working code, but now they've got more complex things to do, that are significantly different than code development.