Parallel Programming and Computer Science 101
It's been said more than once that parallel programming is hard, which suggests that the sooner you get a jump on concurrency, the better off you'll be. So when and how should programmers be introduced to parallalism?According to the folks who teach computer science at the University of Wisconsin-Eau Claire, the answer is right off the bat by integrating parallel programming into all aspects of a student's coursework, rather than treating it as a separate topic. And this means starting in CS1.
To date, three core classes -- CS1, CS2, and Algorithms -- have integrated parallelization. For instance, one of the first topics computer-science students traditionally tackle is algorithm design. At UWEC, students start off with the basics but move into identifying parallelism in programs that involve, say, image convultion, before breaking down parallelization into tasks. They then begin assigning tasks to individual threads. Because these classes are currently taught in Java, the instructors use Java threads and synchronization primitives. But since similar constructs exist in other languages, the examples are easily adapted.
Daniel Ernst and Daniel Stevenson sum up their goals in Concurrent CS: Preparing Students for a Multicore World:
The goal of our curriculum changes is to provide our students with the skills they will need to succeed as we enter the multicore era. By giving our students a clear process for exploiting parallelism and practice at implementing it, they have gained confidence in taking advantage of new hardware. The integration of these topics into standard CS course topics has also caused our students to start thinking about concurrency in all of their programs, not just those given with an explicit directive to parallelize. In a world where every machine is a multiprocessor, we consider it a good trait for our students to be always thinking about ways to take advantage of concurrency.