Impediments to Future Use of Petaflop Class Computers for Large-Scale Scientific/Engineering Applications in U.S. Private Industry

摘要

The environment in government/research HPC sectors is markedly different from that in private industry. Although both involve many of the same applications, the mindset and people/computer resources are significantly different. In this paper, we focus on the barriers to using future HPC machines in the private sector and some current actions and suggestions to overcome these problems. Experts generally agree that the realistic and/or real-time solution of industrial problems will require the use of computers about three orders of magnitude faster than current industrial machines. Impediments discussed include: limitations of ISV-based commercial software, inadequate benchmarking techniques for industrial size problems, limited access to the latest computer architectures and support facilities, paucity of computational science personnel, slow tech transfer of algorithms and modeling techniques from government/research facilities to private industry, and unexplored utilization of Blue Collar Computing™ in private industry.