Evaluating Docker for Teaching and Learning MPI

Abstract

With the increasing development of computer hardware, the performance of computers has increased by using multicores per computer. With the improvement of the multi-core computer design, the teaching of High-performance Computing has become fundamental for Computer Science students to teach them the concept of parallelism. Since the curriculum of teaching HPC does not including setting up a working environment, the teaching of HPC traditionally required access to HPC facility. Moreover, the teaching of HPC required environment reproducibility between teachers and students so that students can reproduce teachers’ environment from the class to run codes in their own time. In addition, students share their environment with teachers so that teachers can marking students' work. In this thesis, we evaluate the use of Docker containers to support the teaching pedagogy of HPC, especially the learning of MPI. The main aims are: (i) provide students with a container that has an MPI compiler and text editor (ii) evaluate Docker for teaching and learning MPI. We explain how we can initiate a Docker image and share it with students using Docker Hub repository; in addition, how students can build their own Docker image and share it with teachers. Furthermore, we evaluate Docker containers in terms of scalability of MPI code on local machine and on Azure Virtual Machine instances.