Novel HPC Parallel Programming Models for Computing (both in CPU and GPU)

Project reference: 2022

As HPC systems get more heterogeneous in nature, it is sometimes feasible to have a single programming model can take a advantage of it. In some cases it is really beneficial where programmers do not need to spend so much time on running some computation when there is enough HPC architecture resource is available.  Novel programming model, for example, Charm++, AdaptiveMPI/Charm++, XcalableMP, Thrust, Kokkos and OmpSS can take a advantage of the computation in heterogeneous architecture platform. It is therefore to analyze their limitation and make a comparison between them on a HPC architecture.

https://sciencebusiness.net/news/eurohpc-partnership-opens-bidding-would-be-supercomputer-hosts

Project Mentor: Dr. Sebastien Varrette

Project Co-mentor: Dr. Ezhilmathi Krishnasamy

Site Co-ordinator: Prof. Pascal Bouvry

Participants: Rafał Felczyński, Ömer Bora Zeybek

Learning Outcomes:

• Learning the novel programming model for the HPC parallel architecture.
• Understanding the limitation between the programming models.
• Better understanding of heterogeneous architecture.

Student Prerequisites (compulsory)
Programming skills in C and C++

Student Prerequisites (desirable):
Basic knowledge in parallel programming model or parallel computer architecture.

Training Materials:

• Kokkos: https://github.com/kokkos/kokkos
• OmpSS: https://pm.bsc.es/ompss
• AdaptiveMPI/Charm++: http://charm.cs.uiuc.edu/research/ampi/
• Thrust: https://thrust.github.io/
• XcalableMP: https://xcalablemp.org/

Workplan:

• Week 1/: Training week
• Week 2/:  Literature Review Preliminary Report (Plan writing)
• Week 3 – 7/: Project Development
• Week 8/: Final Report write-up

Final Product Description:
Benchmarking between the selected list of programming models.

Adapting the Project: Increasing the Difficulty:
Including the low level programming might increase the difficulty in the project, for example, MPI, OpenMP and CUDA programming models and compare against novel HPC programming model.

Adapting the Project: Decreasing the Difficulty:
Benchmarking with fewer test cases.
Comparing the results between just 2 or 3 novel programming models.

Resources:
All of those listed programming models are open source.
Student will get a desktop computer and HPC account at Iris supercomputer of the University of Luxembourg.

Organisation:
University of Luxembourg