High-Performance Computer Science Challenges

John Reynders, Los Alamos National Laboratory

Putting the hardware together to form a TeraScale simulation environment is only a small part of the challenge. Building the software environment to enable applications to run efficiently on such a system is quite another challenge. This talk will outline computer science research under investigation at the Advanced Computing Laboratory at Los Alamos in support of application development for TeraScale platforms. The technologies which will be discussed include scalable run-time systems, object-oriented frameworks, clustering technology, and scalable visualization. In particular, discussion of how these technologies must work together in an integrated fashion to enable requisite performance on clustered shared memory processor systems will be examined. Mixed-model approaches to scalable simulation outside the normal message-passing approach and techniques for the encapsulation of concurrency will be a primary focus of this presentation. These technologies will be described in the context of how they enabled actual applications which scale across thousands of processors in multiple programs at the Advanced Computing Laboratory and how they are aligned with the current procurement strategy of choice for high-end computing in the United States Department of Energy.

Abstract Author(s): John Reynders