Benson Muite, University of Tartu
ISC 2018 BoF on HPC Education: Widening Participation and Increasing Skills through Contests, Challenges and Extra Curricular Learning
This BoF had 4 main speakers, Victor Sande, Matthew Curry, Maciej Szpindler and David McLeod and was moderated by Alexander Ditter. Slides from the presentations can be found here.
After a quick overview of the SigHPC chapter by Benson Muite, the main presentations began.
Student Cluster Competition in South Africa
David McLeod opened the session with an overview of the South African cluster competition. He explained the role of the competition in South Africa’s educational agenda, in particular metrics that are collected and tracked that demonstrate success in attaining the politically determined educational agenda (in particular a skilled workforce that reflects the diversity of the country). David described the challenges of ensuring students with a wide range of skills all learn something. It’s interesting to note that the program has become so successful that it is difficult for all participants to continue in high performance computing within South Africa. Therefore, other research areas may benefit from having participants with good computer science and high performance computing skills.
New Mexico Supercomputer Challenge
Matthew Curry gave an overview of the New Mexico supercomputer challenge. He gave an introduction to New Mexico (a state with below average K12 educational attainment relative to the other states in the USA), and the national laboratories and the universities that are in New Mexico. Issues raised include the role of the supercomputer challenge in building and retaining a strong workforce, as well as financial sustainability of the challenge. One of the main reasons the challenge is successful can be attributed to the efforts of the volunteers (mostly personnel of the national laboratories) involved in mentoring students. Another key component of the program is teacher training, which aims to help teachers teach HPC effectively. Although, collecting and curating metrics for success of the program is challenging, this difficulty enables the program participants to focus on student-focused self-directed learning, rather than just attaining metrics for success. The wide scope of students’ projects makes assigning quantitative measures of performance difficult, thus making it difficult for evaluators to assess them as metrics for success. Post program tracking of student success is also challenging as many high schools do not keep track of their alumni. This program seems best run at a state level.
Educational activities run by the University of Warsaw
Maciej Szpindler gave an overview of activities at the interdisciplinary center for mathematical modeling at the University of Warsaw. Each activity is headed by a different person, and targets slightly different audiences, but enables effective skill transfer. The activities are:
- The workshop for the Polish Children’s Fund,
- Student cluster competitions.
The common aspect of all three activities is to challenge the participants with a difficult problem and support their journey in search of a solution, rather than force them to follow prescribed methods. In other words, the main aim is to support participants in overcoming any technological barriers present in their chosen approach. In case of longer running activities, it is possible to request an external expertise if the approach chosen by participants requires it. The workshop for the Polish Children’s fund is the longest running program and aims to introduce computational science and discovery to high school students. After graduation from high school, a number of these students have gone on to take part in student cluster competitions and mentored other students. Student cluster competitions have received ministerial support to take part in the ISC, ASC and SC cluster competitions.
Spanish Parallel Programming Contest
Victor Sande gave an overview of the Spanish Parallel Programming Contests. This has usually had a small but interested and motivated group of participants. To enable widespread adoption of HPC, the challenge is held in a different location in Spain every year. The goal of the contest is to improve the implementation of various parallel algorithms. It is aimed at university students below 28 years of age, though there is also open participation for those who want to take part but are not interested in wining prizes. The parallelisation is done using OpenMP, MPI and CUDA. A good automated testing and evaluation system called Mooshak has been developed for this contest. This has lead to many improved implementations of parallel algorithms.
Following the presentations, there was a short discussion. The time constraints meant that most of the discussion occurred after the official end of the BoF. The main topics that were raised were the Summer Supercomputing Academy organised by Moscow State University and broader teaching needs being imposed by data science. HPC centers deliver training, that is often individualised, free for participants to attend and does not have the pressure of grading, thus can be much more attractive for students than courses offered by universities. For HPC centers, supporting users is a slightly different activity than teaching students – it is not clear what kinds of provisions should be made for student engagement. There are also differences between teaching/training capabilities of national and local university HPC centers. The material sharing is a good starting point. Attendees also felt that a better user interface design, which currently is a barrier for many existing and potential HPC practitioners, and possibly education on low level programming is required for wider HPC adoption.