DEISA Training Session Jülich, October 23-25, 2006
Venue: Forschungszentrum Juelich GmbH, Juelich, GERMANY
The 2nd DEISA training session was held at Forschungszentrum Juelich GmbH, in Juelich, Germany on October 23rd - 25th, 2006.
30 scientists and experts in supercomputing gathered to discuss topics of HPC and Grid within the DEISA infrastructure. The event started with a general introduction to the DEISA infrastructure, such as the Common Production Environment (CPE) and UNICORE. The special topic of this training session was "Performance and Portability", addressing issues for application developers on improving application performance while keeping optimal portability.
The discussions covered topics on code optimization for large shared-memory clusters, over communication and I/O design to performance analysis with different tools. Supporting the theoretical input from the HPC experts, the participants had the opportunity to experience the DEISA infrastructure fi st-hand in the "Access to DEISA" hands-on session, introduced with this event. Presentations of use cases for the DEISA infrastructure concluded the event, giving insight to recent success stories on improving the scalability of a climate modeling code as well as to visions for collaborative online-visualization utilizing the underlying DEISA infrastructure.
The collected feedback forms prove this training event to be an overall success, creating a positive contribution for the participants' research activities as well as in understanding how the DEISA infrastructure can aid them in achieving their goals.
- DEISA general presentations
- DEISA Overview, Victor Alessandrini, IDRIS, and Achim Streit, FZJ
- DEISA Access, filesystems and job submission, Denis Girou, IDRIS
- DEISA Common Production Environment, Denis Girou, IDRIS
- UNICORE - Introduction, Michael Rambadt, FZJ
- Using UNICORE, Michael Rambadt, FZJ
- The DESHL Client, Malcolm Illingworth, EPCC
- Performance and portability: Tools and techniques for achieving optimal scaling on a cluster of share memory nodes, Gavin Pringle, EPCC
- Performance and portability, use cases
- Scaling ECMWF's IFS Forecast Model to large numbers of processors, George Mozdzynski, ECMWF
- Coupling Applications and Visualization, Wolfgang Frings and Morris Riedel, FZJ