Functional decomposition looks valuable, though as you note, its scalability is often limited. I’ve run into an interesting twist on functional decomposition (perhaps with a note of speculative decomposition in it as well), where people have different algorithms (one example is from image processing) and they don’t know in advance which algorithm will work best on the data in question. So, they just run all the candidate algorithms in parallel and choose the best answer.

As for recursive decomposition, I question whether this is something that typical programmers will readily accept. I know for us computer science types recursion is a powerful tool, but for most of the rest of the planet it doesn’t seem intuitive. Perhaps the right abstractions could help overcome this.

Great outline of various decomposition techniques. I actually never separated Recursive Decomposition and Data Decomposition in my mind, but you are right - they are different, although often used together.

A good special case to mention is Recursive Decomposition with depth of one (1). This one is most commonly used in my opinion.

Best,
Dmitriy Setrakyan
GridGain - Grid Computing Made Simple

The reason: a task, when done, can be leveraged to prune a list of
pending tasks, but this information is only available in real-time; i.e.
one cannot precompute this, rather critical, piece of information
ahead of time.

Now, why is this important? Simply because this is the key when implementing
scalable SAT solvers; in fact, this is the only way to do it