by Alan Beck, editor in chief
San Diego, Calif. — HPCwire conducted an interview with Michael Bernhardt, President & CEO of The Bernhardt Agency, Inc. Bernhardt is a well known strategic marketing and communications consultant who has been active in the HPC industry for 13 years, and a member of the SCxy Industry Advisory Committee. He also serves on a number of corporate and advisory boards for early-growth companies and works closely with several high-technology incubators and Venture Capital (VC) firms.
HPCwire: “What is your perspective on the current state of the HPC industry?”
BERNHARDT: First of all, I think this is a very exciting time in HPC. We’ve seen tremendous progress in applying HPC technologies to vital and important scientific issues in the areas of bioinformatics, life sciences, and weather forecasting, and real-world business applications such as financial, telecommunication, automotive, entertainment, and pharmaceuticals.
We’ve also seen good progress in transferring technology from the government agencies and research institutions to the commercial computing marketplace, resulting in companies like Ford Motor, Charles Schwab, Bristol-Myers, Procter & Gamble, American Express and others using HPC platforms for data mining, data warehousing, and e-business. Many of these companies simply couldn’t be competitive without the use of High Performance Computers.
But, to really appreciate where we are, you have to take a look at where we’ve come from. Just 10 years ago, at SC90 (back then I think it was called Supercomputing ’90) in New York City, Paul Messina led this industry into a new chapter with the formation of the Concurrent Supercomputing Consortium. It was a milestone event for several reasons, but in my mind the public collaboration of so many competing agencies and academic institutions was truly a monumental achievement Over this past decade, companies have come and gone, technology shifts have occurred in both hardware and software, mergers and acquisitions created interesting bedfellows, and collaborations have never been stronger.
Just think about how the HPC landscape has changed. Ten years ago, the largest exhibitors at SC90 were Thinking Machines (TMC) and Alliant…
This year, the largest exhibitors are Compaq, IBM, SGI and SUN.
HPCwire wasn’t even around back then. The publication of record for the supercomputing industry in 1990 was Supercomputing Review, published only in hard copy.
In 1990, Teraflops was a dream…it was beyond the horizon for all of us. Today, we’ve hit that milestone, and are rapidly moving on to the next dream of Petaflops.
HPCwire: “You’re right. We’ve seen a lot of change in this industry. So, how important is the bragging rights to ‘World’s Fastest Computer’ these days?
BERNHARDT: Great question. I don’t think it’s all that important. Over the years, a number of companies have claimed the world’s fastest title, and many times it was based on theoretical peak performance, or theoretical systems that would never actually be built…not based on real machines running production codes. The real measure of leadership in this industry is in the use…or the application of the systems. The HPC industry is made up of a highly intelligent user community and it’s an industry that’s never been enamored with marketing hype. While “world’s fastest computer” has been somewhat of a holy grail for a number of companies, it has no correlation to the long term acceptance of the system, the technology, or the company. More important is where, when and how the systems get used for real-world applications. Most people don’t even remember who actually broke the TFLOPS barrier, at the time…the world’s fastest computer. That milestone was achieved with ASCI Red, an Intel supercomputer built at Sandia National Laboratories. The system was the first to achieve 1 TFLOPS. Using roughly 9,000 standard Intel Pentium Pro processors. I think the Sandia team hit 1.8 TFLOPS right around the time of Supercomputing ’96 and the machine was fully operational the following summer.
A successful vendor in the HPC industry does not need the top performance platform. They have to be competitive of course, but they also must have a product that meets a business need in the commercial market segment (these needs will be vertical market or application specific), and they must also put forth a marketing and sales effort much different from what’s been traditionally used in the HPC market segment.
HPCwire: “So, do you see a killer application on the horizon for the HPC industry?”
BERNHARDT: If you are referring to “killer App” as in the use or application of HPC systems that will be most noteworthy in the near future, I’d have to say it’s the bioinformatics applications for DNA sequencing, genome annotation and analyzing genetic data.
But in terms of the way we often refer to “killer apps”, I’m not sure I see this industry as having a killer application…not like we think of Lotus Notes or Excel…but perhaps a “killer tool” for the HPC industry would be more appropriate. The challenge is to harness the astonishing computing power of these new machines and architectures to achieve higher sustained performance. What’s required is continued advancements in compilers, debuggers and analysis tools that can optimize performance and speed up the application development process.
The exponential growth in companies using high performance computers for product simulation, research, business intelligence, communications, and a host of other applications has resulted in a pressing need for more sophisticated applications that run on multiple platforms and incorporate inordinately complex code. The development of those codes – in a timely and efficient manner – requires application development tools that are intuitive, flexible and adaptable. If you look at the recent surge in use of the TotalView debugger/analyzer and some of the other independent analysis tools and compilers, across all major HPC platforms, you see the proof point for this.
HPCwire: Are you saying that tools are playing a greater role?
BERNHARDT: I’m not sure I’d say tools are necessarily playing a greater role, but they are becoming more important. When you’re running an application across thousands of processors incorporating OpenMP or threads, you need a compiler optimized for the environment, and an analyzer than can locate and correct bugs, analyze load balancing and find bottlenecks in parallel applications.
When you start to use high performance computer systems to run mission or business critical applications like determining the activity, toxicity and absorption of a new drug, or running a massive online trading site, you can’t afford to have glitches in your software.
HPCwire: “It seems like everyone claims their computer is a “supercomputer”, regardless of the size. How would you classify “supercomputers?”
BERNHARDT: Supercomputers are a subset of High Performance Computers. I like the traditional definition of a supercomputer as any computer that is one of the largest, fastest, and most powerful available at a given time. Paul Messina, head of the ASCI program would narrow that down even more and say it’s “a computer that’s among the handful of the most powerful computers in the world for science and engineering.”
I still think of a supercomputer as “Big Iron”…a concentrated computing platform and environment that channels all its power into executing a few programs as fast as possible. Other than that, computers that go beyond the desktop with extremely high performance get lumped together as High Performance Computers.
HPCwire: “What about COTS technology? Is that the real future for HPC?”
BERNHARDT: Yes! This has been a goal for quite some time…the ability to put together a system using commodity hardware components that are clustered together with a common operating language.
And, there’s a good reason for this. To a certain extent, it’s difficult for companies to anticipate future needs. They don’t want to overbuy, yet they need enough computing power to solve their immediate and foreseeable needs. COTS technology enables them to buy what they need now and add on later, with a better assurance that the commodity components will not disappear. It also enables a much wider participation in software and application development.
HPCwire: “Do you see Linux and other open source software being used in more high performance systems?”
BERNHARDT: Absolutely. There’s been a trend towards open source software not so much because it’s less expensive, but because it fosters innovation. Because the code is readily available, product developers can concentrate on both creating unique solutions and freely enhancing the initial code. Linux offers a number of advantages and is now well established within the HPC community.
HPCwire: “Any final thoughts?”
BERNHARDT: The need for more computing cycles is insatiable…we’ll never have enough. The HPC industry will thrive because of this. And on top of that, this is an industry that challenges and pushes itself, an industry that sets monumental goals…an industry of “colleagues” that openly and willingly collaborate for the benefit of science and ultimately the benefit of all mankind. There’s never been a more important time for scientific and technical computing. The one thing that’s guaranteed is more change…and more progress.
Prior to establishing The BERNHARDT Agency in 1994, Mike BERNHARDT was a director for Intel’s Supercomputer Systems Division, with responsibility for strategic marketing communications including public relations, industry analyst relations, product collateral, trade show support, VIP events and user group activities. The BERNHARDT Agency was recently recognized by several independent marketing publications as one of the fastest growing high-technology marketing agencies in the country. BERNHARDT can be contacted at [email protected] .