Method for teaching parallelism on heterogeneous many-core processors using research projects

Parallel programming and parallel architectures are necessary to achieve scalability and performance. It is difficult to evaluate when to teach parallelism and how to change the paradigm from serial to parallel algorithm in traditional curricula. Currently, there are efforts to introduce parallel programming since there are multi-core processors. However, there is a new chip generation called many-core processor. For instance, one processor chip can be built with 1,000 processing cores. Moreover, this type of processor is designed to achieve scalability and performance based on heterogeneous cores. How to teach parallelism to undergraduate and graduate students? Human resources are necessary to design and program parallel architectures based on this next generation of many-core processor. Therefore, the main goal of this paper is to show an experience based on research projects. The idea is to join students from different courses and levels, e.g. Computer Science, Information Systems, Computer Engineering, and Graduate in Informatics. All of them working together in order to understand all characteristics of heterogeneous many-core processors based on integrated environment composed of computer clusters and simulation. The proposed method focuses on projects convergence to teach how to extract characteristics from benchmark traces in order to simulate many-core processors based on Networks-on-Chip. Consequently, students can understand parallel heterogeneous architectures and how to program them. Main results present the number of students interested in this research field along last three years and several scientific papers published. It is important to highlight that papers have students´ participation and two papers are related to education. Results reinforce the contribution of the proposed method since we have several benefits including continuous and cooperative work along more years.