In the fall of 1984, the three of us joined forces to write a proposal to the National Science Foundation. We hoped to bring a supercomputing center to Pittsburgh. The availability of NSF funding to do this was a plum opportunity.
A supercomputing center, we thought, in addition to enabling extraordinary advances in science, could reinforce the research strengths of local universities and industries. It could also facilitate science education at all levels and add to Pittsburgh's presence in computing, helping this region to attract business and retain its exceptional workforce.
- When: April 15, 1 p.m.
- Where: 300 South Craig St.
- What: At DISCOVER11, From Scientific Vision to Scientific Discoveries, PSC will highlight achievements over the past 25 years and present Blacklight, its newest computing resource along with remarks from PSC partners.
- More information: www.psc.edu
On Jan. 17, 1986, after many suspense-filled months, the National Science Board approved funding for the Pittsburgh Supercomputing Center, a joint effort of Carnegie Mellon, the University of Pittsburgh and Westinghouse.
Now that we've been at it for 25 years and built a track record, it seems worthwhile to address some of the questions we know are in the minds of many non-scientists when they think about this somewhat monumental-sounding place, the Pittsburgh Supercomputing Center:
What's a supercomputer anyway? What do you do with them? Why do they need to be so powerful?
And maybe the most important question: Since you as taxpayers make it possible for the Pittsburgh Supercomputing Center to exist, how does what we do make a difference in your life?
Supercomputers -- by common definition -- are the most powerful computers available at a given time for scientific computing. As you might imagine, this field is dynamic, and our center has participated in breathtaking technological change.
Supercomputer performance is usually measured in flops -- floating-point operations per second. Basically, that's the number of arithmetic operations -- adds and multiplies, etc. -- you can do in a second.
Our first supercomputer, a CRAY X-MP, could do 840 million flops. Today's supercomputers are 100,000 times more powerful and cost-effective.
A 2011 laptop has more computing power than that CRAY X-MP -- which cost $18 million. Because of this extraordinary technological evolution, we can tackle problems now that we didn't dream about 25 years ago.
For our first decade, supercomputers employed a highly specialized technology. In the 1990s, however, they began using the technologies of home computer and business systems, exploiting the economies of this large market. To attain their enormous processing power, they emphasized teamwork -- linking processors so they could work together efficiently, technically called parallelism.
Today parallelism has progressed to where the most powerful systems have as many as 100,000 processors working together, with performance in trillions of flops.
Why do scientists, engineers and a broadening host of commercial enterprises need these ever increasing amounts of computing? For many scientific problems with great societal impact, what we have today has been very productive but doesn't take us as far toward solving these important problems as we'd like.
A prominent example is climate modeling. How can we reliably know whether human activities that add carbon dioxide to the atmosphere are changing the Earth's climate? You can't do experiments at the scale of the whole planet. Computational models that we can rely on within statistically defined levels of uncertainty are crucial to understanding this problem.
These models are remarkably complex, building in all the phenomena that affect climate -- structure of the atmosphere, wind currents, surface topography, the warming effect of sunlight, ocean currents, salinity and temperature, the insulating effect of clouds and many other factors.
Reliability of these models depends on how finely detailed they are. Ideally, for example, they would include every tree and building on the planet. The catch, though, is that as a model focuses more sharply on fine detail, computing requirements expand. Computational scientists spend a good portion of their brain cycles to arrive at the optimum balance between realism in their models and available computing power.
We have a system -- called Anton -- that exemplifies another instance in which it takes incredible amounts of computing to solve problems that affect our lives. Since Watson and Crick famously deduced the helical structure of DNA, molecular biology has been a major effort of science, with great amounts of energy focused on finding the structure and function of proteins, the chemical building blocks of which we are made.
The payoff -- along with intimate knowledge of processes at the core of life -- is using this knowledge to develop new therapeutic drugs with reduced side effects and greater ability to defeat disease.
Unlike their static pictures in biology textbooks, proteins in their living environments are protean -- constantly changing shape. These fast, jiggling, dance-like movements -- which are key to protein function -- can be "seen" only through a computational technique called molecular dynamics.
Anton (named for Anton van Leeuwenhoek, the father of microscopy) is a "computational microscope" that can look more deeply into the sub-microscopic realms where proteins do their work.
Built and donated to our biomedical group for a year, by D. E. Shaw Research in New York City, because of our group's experience in biomedical supercomputing, and funded through the National Institutes of Health, Anton is the world's most effective system for this kind of work. It is specialized to run only molecular dynamics, runs it 100 times faster than other supercomputers, and therefore can "see" the motion of proteins over longer and more meaningful time spans.
Nevertheless, there are many important protein motions that extend in time beyond Anton's ability to see and that await further technological advances.
Our newest supercomputer, Blacklight, is notable -- along with its speed -- for its huge amount of easily accessible memory: 32 trillion bytes of "shared memory."
Shared memory means all of a system's memory can be accessed directly from all of its processors, making it -- relatively speaking -- much easier to program and use, and useful for research that requires large amounts of data to be readily available in memory. Such projects include large database-searches, such as recent research at Carnegie Mellon that demonstrated how information on social-networking web-sites and other online sources can risk the privacy of Social Security numbers.
Blacklight is the largest shared-memory system in the world. It became available this year as part of the TeraGrid, the NSF program that links computational resources at 11 centers across the country.
Along with our role in U.S. computational science and engineering, the Pittsburgh Supercomputing Center provides Internet and educational services to this region. Our networking group operates a high-speed Internet hub that connects CMU, Pitt, Penn State and other universities and public school districts to national research and education resources. Last year our network staff led a coalition of state organizations in an effort that won $100 million in federal stimulus money to build a high-bandwidth network across Pennsylvania. Construction of this network, which will serve businesses and medical centers and improve rural access to the Internet, is now under way.
With a recognized need nationally to improve education in science, technology, engineering and math, our center carries out several programs aimed at helping to create a cyber-savvy workforce and science-literate citizens. Our staff has trained many southwestern Pennsylvania high school teachers in easy-to-use modeling and simulation tools that bring "cool" technology into the classroom.
We also introduce high-school teachers to bioinformatics -- using computers to search and analyze genes and other biological data -- which as a result has become a course in several regional high schools. Through a program called SAFE-Net, we provide materials for learning about cyber threats and questions of cyber ethics that arise from social networking and other uses of the Internet. Our biomedical group provides multidisciplinary teaching materials as online modules and DVDs in biologically related topics such as diffusion and osmosis.
From a national perspective, our prime measure of success is scientific accomplishment and its practical benefits. Researchers who use our resources -- numbering in the thousands each year -- are from government and industry as well as hundreds of universities. Availability of our systems, and as importantly, our pool of computing professionals, has led to thousands of published scientific papers.
Among this work is heart modeling that guided development of a practical prosthetic heart valve. Protein simulations at our center that explain how a biologically crucial channel lets water into cells were cited in the 2003 Nobel Prize for chemistry. This knowledge advances research on diabetes and kidney disease.
In 2005, storm-forecast modeling we've collaborated on accomplished a first in weather forecasting by successfully predicting location and structure of a severe thunderstorm a day in advance. As this advanced technology, developed with supercomputing, is adopted in the National Weather Service and other forecast systems, such as commercial airlines, it will save billions of dollars and many lives.
Industrial work at our center has included beverage-can modeling by ALCOA, finding designs that use the least possible aluminum for needed strength. Quantum simulation of coatings for sunglasses with "transition lenses" for PPG Industries has saved money for PPG and reduced the time to market for a new product.
During the 2009 H1N1 flu outbreak, epidemiological modeling at our center, in collaboration with Pitt's Graduate School of Public Health, supported decision makers in Allegheny County and in Washington and led to new thinking about the mitigating effect of school closing. Recently, our scientists co-authored a paper in Nature, the prestigious international science journal, that for the first time traces connections among neurons in a portion of the visual cortex, a first step toward a major scientific goal -- a "wiring diagram" of the brain.
Partnerships have been essential, primary among them our collaboration with Westinghouse. Rather than building a facility from scratch, we were able to leverage their infrastructure, locating our supercomputers at the the Westinghouse Energy Center in Monroeville, from where they're linked to our administrative offices in Oakland by high-speed fiber.
Our association with CMU and Pitt has been extremely important. Our center was one of the first major research efforts to receive official backing from both Oakland-based universities. Affiliation with two great universities has helped in many ways, among them attracting talented people. Currently numbering about 90, our staff are an exceptional reservoir of talent and experience in supercomputing.
We're lucky to work with them and also pleased to be, in a sense, voyeurs -- to see how people are making important scientific accomplishments happen because powerful tools for revealing knowledge about our world are progressing at an astounding pace.
To be a computational scientist is to live in very interesting times.