Using Patterns to Teach Parallel Computing

In this paper, we describe the results of teaching a parallel programming course using a pattern programming approach in a course taught across the State of North Carolina on a televideo network in Fall 2013. Five universities participated in this study. The course begins with a higher-level tool called the Seeds framework that creates and executes high-level message passing patterns such as a workpool without writing low level MPI code. To avoid going directly to MPI next, we used another tool (Paraguin compiler) which uses compiler directives to create MPI code for patterns. Once students understand the pattern programming approach we then present low level MPI routines and their more complex parameters but now with the knowledge of parallel patterns. An independent professional evaluator is employed to deploy survey instruments and produce an analysis of the results. The lessons we learned from this data collected in Fall 2013 are: 1) Teaching parallel computing in the context of patterns has a positive impact on student learning, 2) Teaching the lower level tools first would be beneficial, 3) The improvements made to the Paraguin compiler directives significantly improved the students confidence in using the tool, and 4) The lower level tools can still be taught in the context of patterns.