Abstract:

In Computer Engineering degree courses that deal with High-Performance Computing, traditionally linked to the branch of Computer Architecture and Technology, students tend to work in languages such as C, libraries such as MPI, and focus their practice on accelerating calculations. However, this may limit students' vision for the design of high-performance systems in other equally relevant areas and based on alternative technologies. Although these contents can be taught theoretically, their guided practical visualization could reinforce the acquisition of fundamental competencies. This work designs a complementary activity to show how parallel computing can be determinant in developing IoT solutions using the MQTT communication protocol, Python, and Deep Learning. The activity is based on increasing the number of processable messages per unit of time with this protocol when processing is computationally expensive. To this end, we work from the sequential version to different parallel designs, including the communication of processes using shared files and the creation of RAM units. This activity has been planned as a seminar within the High-Performance Computing and Architecture course of the Degree in Computer Engineering of the University of Granada. However, it can be implemented in equivalent subjects of other curricula. As a seminar, it has a theoretical orientation with eminently practical tasks integrated. The timing can be adjusted between 2 and 6 hours, managing the codes provided totally or partially by the teacher. This work shares the sample code developed and the results obtained as a reference. This way, the proposed activity is easier to adopt by other professors with similar interests.

Keywords:

Teaching, Parallelism, High-Performance Computing, MQTT, Deep Learning, Python.