PRACE Summer School: Taking the Most Out of Supercomputers
In the PRACE Summer School the participants will learn advanced programming skills, which are necessary for getting the most out of the largest ("Tier-0") supercomputers.
The school will cover the following topics, addressed with lectures and related hands-on exercise ("lab") sessions.
- William Gropp (University of Illinois Urbana-Champaign): Advanced features of the message-passing interface (4 h lectures, 3 h labs)
- Rolf Rabenseifner (HLRS) & Pekka Manninen (CSC): Hybrid message passing + shared memory programming (3 h lectures + 1 h lab)
- Sebastian von Alfthan (FMI) & Sami Ilvonen (CSC): Using GPGPUs in high-performance systems - CUDA+MPI programming (3 h lectures, 2 h labs)
- Brian Wylie (Forschungszentrum Jülich): Performance analysis tools for massively parallel applications (2 h lectures, 1 h lab)
- Jussi Enkovaara (CSC): Using Python in high-performance computing (1 h lecture)
- Jan Westerholm (Åbo Akademi): Low-level (assembler) serial code optimization (1 h lecture, 1 h lab)
In addition to these, four scientific case studies concerning using the top-end HPC systems to solve the most challenging problems will be presented by European top researchers:
- Lennart Johnsson (KTH and University of Houston): Towards exascales
- Erik Lindahl (KTH): Molecular simulation - paths to exascale
- Kai Nordlund (University of Helsinki): Scaling a molecular dynamics code to simulate billions of atoms on one hundred thousand processors
- Kari Rummukainen (University of Helsinki): "New physics" with supercomputers
The exercises will be carried out on a brand new Cray XE6 system in PDC, Sweden as well as on CSC's Cray XT4/XT5 system Louhi.
The attendees are encouraged to bring a poster on some scientific application of large-scale HPC or on HPC methodology development.
Prerequisites
Intermediate skills and experience in parallel programming, e.g. skills equivalent to those addressed in the CSC Summer School.
Program
| Mon Aug 29 |
Tue Aug 30 |
Wed Aug 31 |
Thu Sep 1 |
|
| 9.00-9.45 | Rummukainen keynote |
CUDA Lec I | Tools Lec I | |
| 9.45-10.30 | Advanced MPI Lec I | CUDA Lec II | Tools Lec II | |
| 10.30-10.45 | Opening | Coffee break |
Coffee break |
Coffee break |
| 10.45-11.45 | Johnsson Keynote |
Advanced MPI Lec II |
Lindahl Keynote | Tools/optimization Lab |
| 11.45-12.45 | Lunch break |
Lunch break |
Lunch break |
Lunch break |
| 12.45-13.30 | Hybrid Programming Lec I | Advanced MPI Lab I | CUDA Lec III | Nordlund Keynote |
| 13.30-14.15 | Hybrid Programming Lec II | Advanced MPI Lab II | CUDA Lab I | Keynote cont'd + closing |
| 14.15-14.30 | Coffee break |
Coffee break |
Coffee break |
|
| 14.30-15.15 | Hybrid Programming Lec III | Advanced MPI Lec III | CUDA Lab II | |
| 15.15-16.00 | Hybrid Programming Lab |
Advanced MPI Lec IV |
Low-level Code Optimization |
|
| 16.00-16.15 | Break | Break | Free | |
| 16.15-17.00 | Python in HPC |
Advanced MPI Lab III |
Free | |
| 17.00- | Welcoming drinks |
Poster session |
Dinner |
Materials
- Agenda and practicalities
- Instructions on Lindgren system
- Hybrid MPI+OpenMP programming tutorial
- Advanced MPI tutorial
- CUDA+MPI tutorial
- Scalasca tutorial
- Python in HPC
- Low-level optimization
Registration
Additional information
- See recommendations and special rates for accommodation options.
- See how to find to the CSC's premises.
- Send all inquiries to courses(at)csc.fi.
- The school is jointly organized by CSC and SNIC-KTH within the FP7 project PRACE-1IP. For more information about the PRACE-1IP project, see www.prace-project.eu.
