39th SPEEDUP Workshop on High-Performance Computing

ETH Zurich, Switzerland,
September 6 - 7 2010

Sponsored by: Speedup         D-INFK  


This workshop's focus: Parallel software environments and fault tolerance

The intention of this workshop is to present and discuss the state-of-the-art in high-performance and parallel scientific computing. Presentations will focus on algorithms, applications, and software issues related to high-performance parallel computing. The focus of the workshop on Monday Sept 6 will be on software environments for large scale simulations and on the issues of fault tolerance in massively parallel systems.

The scientific program of Sept 6 consists of seven 45-minute talks and a poster session. Please encourage your collaborators to uploaded an abstract. The deadline is August 27, 2010.

On Sept 7 we will organize a tutorial on hybrid MPI/OpenMP computing. It will be taught by Sadaf Alam and Neil Stringfellow (both CSCS, Manno, Switzerland).

Directions

The conference will take place in the CAB building of the ETH, lecture hall G11.

Program

Monday, September 6

8:45 - 9:00
Welcome
9:00 - 9:45
Petascale AMR, with applications to solid earth geophysics problems
Omar Ghattas (University of Texas, Austin)
9:45 - 10:30
PETSc: the Portable, Extensible Toolkit for Scientific Computation
Matthew Knepley (Argonne National Laboratory)
10:30 - 11:00
Coffee break
11:00 - 11:45
Trilinos for Emerging Parallel Computing Systems
Christopher Baker (Oak Ridge National Laboratory)
11:45 - 13:15
Lunch break
13:15 - 14:00
UG (Unstructured Grids): Development of PDE Solvers for Biomechanics
Rolf Krause (USI Lugano)
14:00 - 14:45
Fault tolerance issues in large scale applications
Romain Teyssier (University of Zurich)
14:45 - 15:15
Coffee break
15:15 - 16:00
Beyond Application-Level Checkpoint/Restart - Advanced Software Approaches for Fault Resilience
Christian Engelmann (Oak Ridge National Laboratory)
16:00 - 16:45
Hybrid MPI/OpenMP computing
Georg Hager (University of Erlangen-Nuernberg)
16:45 - 17:30
Apero
14:45 - 17:30
Poster Session
17:30
General assembly of the Speedup Society

Tuesday, September 7

Tutorial on hybrid MPI/OpenMP computing

Sadaf Alam and Neil Stringfellow (CSCS, Manno, Switzerland)

08:30 - 09:00
Historical Overview
The rapidly changing nature of HPC architectures and the programming models used to develop applications on them, the multitude of library and directive implementations and the standardisation on MPI and OpenMP.
09:00 - 10:00
Why MPI/OpenMP?
The main reasons for choosing MPI/OpenMP hybrid programming model. The architecture of modern multi-core multi-socket clusters and MPP systems. The advantages of a combined message passing and shared memory threaded approach for parallel execution. The approaches available in MPI and OpenMP for process distribution and thread level parallelism, the use of worksharing constructs and task parallelism. A comparison of MPI/OpenMP with MPI/Pthreads approaches.
10:00 - 10:30
Coffee Break
10:30 - 11:00
Examples
of hybrid parallelisation using simple test codes, and discussion of improved resource utilisation that can be achieved
11:00 - 12:00
Performance issues
with MPI/OpenMP hybrid parallelism. Thread initialisation and management overheads. Thread synchronisation. CPU affinity, memory affinity, cross-socket thread groups. Allocation /initialisation of memory for thread groups.
12:00 - 13:15
Lunch
13:15 - 14:00
Compilers, libraries and tools
Tools for process and thread placement on clusters and MPP systems, use of threaded libraries, load balancing, optimisation restrictions imposed by compiler directives, re-engineering algorithms for improved threaded performance. Status of MPI and OpenMP implementations, performance results.

Invited speakers

14:00 - 14:30
TAU performance analysis tool for tuning and optimisation of MPI/OpenMP hybrid applications
Professor Allen Malony, University of Oregon.
14:30 - 15:00
User experiences 1 - The FISH astrophysics code for modelling supernova explosions
Mr Roger Kaepelli, University of Basel.
15:00 - 15:30
Coffee Break
15:30 - 16:00
User experiences 2 - Hybrid parallelisation of the BigDFT code
Professor Stefan Goedecker, University of Basel.
16:00 - 16:30
User experiences 3 - MPI/OpenMP implementation within CP2K, scaling to tens of thousand of cores
Dr. Joost VandeVondele, University of Zurich.
16:30 - 17:00
Future directions for MPI and OpenMP in the massively multicore era.

There is a small fee of CHF50 payable at the workshop. For students the workshop is free of charge. Details and the registration form can be found here.

The fee for the tutorial are CHF150 payable at the workshop or tutorial. For students the price is CHF50.

Organization and program committee