The Council on Competitiveness is a consortium of industry and academic leaders focused on continuing to secure the competitive economic position of the United States. As part of these activities, the Council coordinates a high-performance computing (HPC) initiative; I am a member of the initiative's advisory committee. The Council has sponsored HPC surveys on users and software, and it and released a video (targeting a non-technical audience) on the economic benefits of HPC.
The surveys identified two primary impediments to broader commercial uptake of HPC: (a) the lack of trained computational science personnel and (b) the small number of commercial codes for today’s parallel systems. The lack of computational scientists is a national issue of rising importance, especially as computing is increasingly the critical enabler of scientific discovery. When I chaired the PITAC subcommittee on computational science, we debated many possible approaches to training more computational scientists, all of which seem to require changes to university degree programs and structures. The 2005 report outlines these issues.
Computing for science is especially challenging because an effective computational scientist is facile in multiple technical domains, ranging from algorithms, software and architecture to numerical analysis, physics, chemistry, environmental science and biology. All too often, our educational programs make it extraordinarily difficult for a student to navigate the chains of educational prerequisites needed to acquire these diverse skills.
The lack of robust, turnkey tools from independent software vendors (ISVs) is a consequence of the HPC transition from supercomputers based on custom vector processors to large-scale parallelism based on commodity microprocessors. One can trace the rapid technology shift in the demographics of the Top 500 list. Why is this a problem? I’m glad you asked!
Large scientific and engineering applications, whether for finite element codes for crash simulation or mesoscale weather modeling, are developed over a decade or more and often contain tens or even hundreds of millions of lines of code. The historical attraction of vector systems was their high delivered performance for sequential codes when automatically vectorized by a compiler.
Today’s distributed memory parallel systems, though delivering phenomenal peak hardware performance, but the applications must be laboriously parallelized manually if they are to achieve any reasonable fraction of this peak. Porting and optimizing such codes on diverse and changing computer architectures is a daunting task, particularly for small companies. My friend Vince Scarfino from Ford commented on these difficulties in his 2003 testimony to the House Science Committee.
Despite these difficulties, there is no doubt that HPC is an enormous competitive advantage for U.S. companies – if it can be deployed effectively. One need look no further than Proctor and Gamble’s analysis of Pringle aerodynamics for production automation, Boeing’s design of next-generation aircraft, or American Airlines logistics planning based on high resolution weather forecasts. The challenge is to deliver this advantage more broadly, to companies large and small in diverse technical domains.
With this backdrop, last week I participated in a meeting hosted by the Council on HPC as an enabler of economic prosperity. In addition to my compatriots from the Ohio Supercomputer Center (OSC) and the Pittsburgh Supercomputing Center (PSC), there were representatives of the Economic Development Administration (EDA) from the Department of Commerce and the NIST Manufacturing Extension Partnership (MEP). The goal of the meeting was to exchange ideas about state and regional economic development and use of advanced technology for global competitiveness.
At the meeting, I described RENCI’s model of a virtual organization, with an anchor site and distributed engagement sites that work collaboratively to address statewide problems, from disaster response and its economic impact through rural health care to economic data mining and broadening the base of participation in science and technology. This virtual organization model generated considerable interest and discussion, based on current national approaches to economic development via regional partnerships that strike a balance between city and state foci.
In addition to my presentation, OSC Director Stan Ahalt described their Blue Collar Computing partnerships and collaborations and Beverly Clayton described PSC’s industrial partnerships. A common theme across all three summaries was the need for engaged partnership, where teams from the center and the company work collaboratively to address strategic company problems.
In the subsequent discussion, I noted that in my experience companies, large and small, most benefit from intellectual engagement addressing strategic challenges using a variety of technologies. As I put it colloquially, “As a CEO, what keeps you awake at night, worrying about disruption to your market niche or challenges to your five year competitive position?” This engaged approach was successful in our Private Sector Program (PSP) when I was at NCSA.
We are tuning the approach in North Carolina to help address a changing economic climate, the demise of traditional industries (furniture, textiles and tobacco) and rapid globalization. Simply put, partnering with North Carolina companies to ensure their global competitiveness is part of RENCI’s state mission, using HPC and a variety of other technologies.