May 7, 2006
[ return to the news list ]
HPC Wire (04/07/06) Vol. 15, No. 14,
In a recent interview, the Defense Department's Douglass Post outlined his thoughts on DARPA's High Productivity Computer Systems (HPCS) program, which has awarded funding to Cray, IBM, and Sun Microsystems to advance supercomputing to the petaflop level while delivering systems that are simple to use and program. The program aims to cut the time-to-solution for both code development and production, while simplifying the complexity of the computer's architecture to take advantage of increased power. The vendors are trying to improve on the traditional Linux cluster to create hardware and software for better floating-point and integer-arithmetic computations. Productivity is a key part of the project, Post says, and the developers are keeping the flops/dollar metric in mind as they create applications that address real problems. A dedicated productivity team is conducting case studies to identify bottlenecks in the development process, and the vendors are creating new languages and tools to express highly abstract parallelism. The project is also working to consolidate the languages being developed by the vendors to create a single language that will be adopted by the community. This summer, DARPA will select one or two vendors to receive funding to bring a multi-petaflop computer to market by 2010. MPI is a workable language for supercomputing, Post says, though there is a vast potential market for a version that can perform at a higher level of abstraction. In developing new languages, the vendors are keeping market acceptance a priority, and are working to ensure that they operate on multiple platforms, while the Argonne National Laboratory is leading the effort to consolidate the research into a single language.
For the full article see http://www.hpcwire.com/hpc/614665.html
Posted by rab5 at 07:57 PM
[ return to the news list ]
