Ken Miura described briefly the Next Generation Supercomputer project in Japan. It is a 7-year project running from 2005-2012 with a budget of 110 billion Yen. All three main computer manufacturers of Japan, Fujitsu, Hitachi, and NEC, are involved in the systems' development and the entire project is overseen by RIKEN, the national technical research institute that also guided the development of the Earth Simulator system. Interestingly, from the very start the application developers are involved: 21 applications are selected which are expected to influence the ultimate system architecture. A system is expected to be in operation by 2011 with further extension and tuning in 2012. All selected application are expected to perform at the Pflop/s level by this time.

Five panelists were selected that gave position statements with respect to Petaflops computing. Pete Beckman, from Argonne National Lab, detailed some requirements for Petaflop systems to be useful: there must be a proper hierarchy in the OS components because of the distinct parts that will make up the system, like management nodes, compute nodes, and I/O nodes. Also much has to be done about the power requirements: it is important to bring down the power use to about 1 Gflop/Watt when one wants to have a practically usable system. Yasuhiro Idomura, JAEA, explained his need for Petaflop systems for doing 5-D and 6-D plasma fusion simulation in a full Tokamak geometry, while Thomas Lippert from Juelich Research Centre did the same for nanostructure research that now is done on the current machines in Juelich but falls short on detail and extent for biomolecules, a main topic of interest.

Steve Scott, Cray Inc., sketched the road to Petaflop systems as visioned at Cray's. He also subscribed to the idea of a heterogeneous system because no single processor type is optimal for all algorithms/operations within real-life applications. A first Petaflop (peak) system will be delivered to ORNL in 2008 and Scott expected to reach Petaflop sustained performance by 2010.

Lastly, Adrian Simmons from ECMWF described the need and benefits of more detailed weather/climate models in view of more extreme weather phenomena and details of global warming. Presently they are using a 2112 CPU IBM POWER5 system but the higher detail comes with a growth factor of N^3 of N^4. So, even a fourfold higher resolution would require systems that are essentialy unattainable at the moment.

Unfortunately, the postion statements of the panelists, together with the introductions of the chairmen took so much time that no time was left at all for discussion between the panelists and with the public, which is normally the purpose of a panel session. So, although the presentations were mostly interesting in themselves, there was definitely something missing in this part of a generally interesting conference.