Next Generation Software: A Framework For Developing Complex Applications On High-End Petaflop-Class Machines

University Of Delaware, Newark DE

Investigators

Abstract

0103723-University of Delaware-Guang R. Gao A Framework for Developing Complex Applications On High-End Petaflop-Class Machines This research proposes to develop novel software technologies for supporting the design and implementation of complex applications on next-generation high-performance computers such as the IBM Blue Gene machine, the teraflop-class systems of the ASCI program, and the architecture (s) proposed in the NSF peta-flops point-design studies. Future high-performance computers such as these will have large numbers of processors (from a few thousand to several hundred thousands, or even more), and a complex multilevel memory hierarchy with memories physically distributed across different parts of the machine. To use such machines effectively, enormous amounts of parallelism must be exposed in user programs, and careful attention must be paid to the latency and bandwidth of access to different levels of the memory hierarchy. The applications enabled by such high-end machines are also expected to be significantly more complex and dynamic than applications in the past. Therefore, supporting the development of such applications on next-generation high-end computers is a major software challenge.