Where does Asia stand? This rising supercomputing power is reaching for real-world HPC leadership.
The United States has lost the nominal supercomputing title twice--both times in the past decade to an Asian country. I say "nominal," because the most widely recognized title is the top spot on the semi-annual list of the world's Top500 supercomputers (www.top500.org). The Top 500 list is a valuable census of big supercomputers, but its single benchmark, High Performance Linpack, was never to predict which supercomputers would perform best on a spectrum of real-world applications.
As of this writing, the real-world speed record is held by Oak Ridge National laboratory's "Jaguar." To IDC's knowledge, this is the only supercomputer so far to crack the petaflop barrier ([10.sup.15] operations/second) on a complete, 64-bit user application.
Still, by taking the lead on the Top 500 list, a supercomputer shows it has the potential to seize the number-one spot on real-world applications down the road. That's why the world took note when Japan, and then China, vaulted to the number one position on the list.
* First Asian World Champion: Japan--From 2002 to 2004, Japan's "Earth Simulator" supercomputer topped the Top 500 list, sparking concern that motivated the U.S. Government to substantially bump up funding for high performance computing in order to recapture the lead.
* Second Asian World Champion: People's Republic of China--In November 2010, China seized the number one spot on the Top 500 list, becoming the second Asian nation to wrest nominal leadership from the U.S.
In his 2011 State of the Union ad dress, President Obama noted China's achievement and countered that the Department of Energy's Oak Ridge National Laboratory is "using supercomputers to get a lot more power out of our nuclear facilities."
Will the U.S. Government respond to China's supercomputing achievement, as it did to Japan's, by seriously bumping up HPC funding? In the midst of Washington's current budget-cutting fervor, this is far from certain.
Why is HPC important for national economies? Because supercomputing has been firmly linked to economic competitiveness, as well as scientific advances. In one IDC study, 97 percent of companies that had adopted supercomputing said they could no longer compete or survive without it. To out-compute is to out-compete.
Politicians increasingly see this.
* In his 2006 Stare of the Union address, President George W. Bush promised to trim the federal budget, yet urged more money for supercomputing.
* In 2009, Russian President Dmitry Medvedev warned that, without more investment in supercomputer technology, Russian products "will not be competitive or of interest to potential buyers."
* In June 2010, Rep. Chung Doo-un of South Korea's Grand National Party echoed that warning: "Il Korea is to survive in this increasingly competitive world, it must not neglect nurturing the supercomputer industry, which has emerged as a new growth driver in advanced countries."
* Europe's PRACE supercomputing program is opening its arms to industry as well as science.
TODAY'S HPC SCENE IN ASIA
Japan and China captured nominal supercomputing leadership with their Top 500 achievements. Where do Asia's major HPC user-countries stand in relation to real-world HPC leadership? A daunting question, but I'll give it a try.
People's Republic of China
China has employed scientific computing since the 1950s. By the mid-1990s, the China Meteorological Administration was using supercomputers from Cray and IBM, as well as China's own Shenwei-1 system, for numerical weather forecasting. China's petroleum industry also has long-standing HPC credentials.
These and other HPC experiences enabled China to build a strong cadre of computational scientists and engineers. Through exposure to U.S. systems and its own, China also gained experience in designing supercomputers.
Long deprived of access to the most powerful class of foreign supercomputers for political reasons, China now plans to develop its own HPC technology from the ground up, including the processors. The Tianhe-1A supercomputer that scaled the summit of the Top 500 list in November 2010 made a start in this direction with home-grown software technologies and a custom ASIC processor.
The first order of business for China is to expand its domestic HPC market.
"China is developing a significant internal ecosystem for HPC," according to Dr. David Kahaner, founding director of the Asian Technology Information Program (ATIP). "It's a country with one billion people, a fast-growing economy, and national pride."
Kahaner said that, in 2010, Chinese HPC vendor Dawning outsold IBM in China for the first time, and 50 percent of the country's HPC system sales went to Chinese vendors.
This is not to imply that China is closing off its burgeoning economy to foreign competition. In 2010, for the first time in its 102-year history, General Motors sold more vehicles in a foreign country--China--than in the U.S.
Less well-publicized than Tianhe-1A is the fact that China's government also deployed five public HPC cloud computing centers recently, with free access for industry. The government views cloud computing as a business model, a way to make HPC access ubiquitous in the absence of a pervasive network of traditional HPC data centers. This is analogous to the preference for cell phones in countries lacking strong wired telecommunications networks.
According to David Kahaner, if the Chinese government succeeds in using cloud computing to push HPC down into SMBs and the end-to-end manufacturing process--design, manufacturing and logistics--that could further turbo-charge China's robust economy. IDC research shows that, today, Japan, major European countries and the U.S. are well ahead of China in using HPC within the supply chain.
Japan's economy and HPC market are both far more mature than China's. For decades, Japan vied with the U.S. for HPC leadership--which Japan held from 2002 to 2004. Japan has contributed enormously to the advancement of computational science and HPC technology. And no country has gone farther than Japan in applying HPC to major industries, from automotive to telecommunications and consumer products.
Japan's HPC system vendors--Fujitsu, Hitachi and NEC--were late in switching from custom vector processors to standard microprocessors. This led to retrenchment in non-Japanese markets where Japan had been strong, especially Europe. On top of this, Japan's economy has been mired for years in a recession. In September 2009, the Japanese Government had to rescue HPC from a budget-cutting panel's proposal that would have slashed funding for supercomputing projects to almost nothing. The government reduced the supercomputing budget only slightly--a strong vote of confidence in the importance of HPC.
Before this showdown, Fujitsu, Hitachi and NEC were collaborating on the nation's next-generation supercomputer project, a hybrid vector-scalar HPC system nicknamed "Keisoku." In May 2009, NEC and Hitachi exited the project, leaving Fujitsu as the last man standing. The 10-petaflop "Keisoku" supercomputer, slated to become operational in 2012, will now be based on scalar processors from Fujitsu. The good news: dropping the vector component lowered the development challenges, and a single-architecture system should be a leaner, meaner champion for Japan in the multinational race for HPC leadership.
With the global HPC market back in a healthy growth mode, especially at the high end, Fujitsu has announced plans to re-enter the European market. NEC and Hitachi also have deep HPC experience and technical muscle. No one should count Japan out of the HPC leadership race.
Republic of Korea
The Korean Government also recognizes the competitive value of HPC. As noted earlier, Rep. Chung Doo-un called for a systematic plan to boost investment in HPC technology and HPC systems.
South Korea has a strong HPC history, including at the high end. The Korea Meteorological Administration (KMA), for example, has long been one of the world's foremost centers for numerical weather prediction (NWP) and climate research. The combined 754 peak teraflops of the center's two Cray XE6 supercomputers arguably make KMA the world's most powerful facility for operational NWP.
Through IDC's experience holding HPC User Forum meetings at the Indian Institute of Technology (New Delhi) and the Indian Institute of Science (Bangalore), along with market tracking, we have a good sense of government funding patterns. For the most part, India has opted to fund multiple small to midrange HPC centers around the country, rather than concentrating on one or more large centers. This approach may reflect political realities as much as budgetary constraints. As a consequence of this policy, there are relatively few large-scale supercomputers in India today--four on the November 2010 Top 500 list, compared with China's 41.
India's most powerful supercomputer, as of that date, was "EKA," a 172-teraflop HP cluster installed at Tata Sons' Computational Research Laboratories in Pune. The runner-up was a 70-teraflop IBM Power system at the India Institute of Tropical Meteorology, and third was a 54-teraflop version of India's own "PARAM Yuva" cluster. Like China, India for years was not permitted to acquire large U.S. supercomputers. In response, India developed its own HPC technology.
India is an HPC software powerhouse and has world-class expertise in many HPC-supported scientific and engineering domains--including severe storm prediction and digital content creation.
Elsewhere in Asia
Singapore and Hong Kong also bare supercomputers on the November 2010 Top 500 list. Taiwan recently announced plans to deploy a 170-teraflop cluster at the National Center for High Performance Computing. (Australia's vibrant HPC culture isn't covered here and deserves separate treatment.)
WHAT DOES IT MEAN?
In the formative HPC market of the mid-1980s, competition was between hot "muscle cars," proprietary supercomputers from U.S. and Japanese vendors who were forced to innovate heavily, because there were few standard technologies to buy in the open market.
Thanks to the steady march of commoditization, today "it's not the size of your muscles that counts" (as a successful Arnold Schwarzenegger told a young bodybuilder), "it's the size of your wallet." Commoditization allows more nations and regions to enter the global race for HPC leadership, assuming they have the money and willpower to fund big supercomputers.
In most cases, funding means purchasing HPC systems already available in the marketplace. In other cases, including China and India, it also means funding R&D to develop home-grown HPC technologies. Asia will not be, as it once was, simply a market for U.S. and Japanese supercomputers.
Asia is a rising HPC power. The ultimate measures of supercomputing leadership are HPC-enabled scientific and engineering advances--what happens when creative minds run real-world applications. From a competitive standpoint, Asia's growing lead in graduating Ph.D. scientists and engineers should concern the rest of the world more than who's on top of the Top 500 list. It will be interesting to see what innovations Asia's user communities achieve in coming years with stepped-up HPC capabilities.
Steve Conway is Research VP, HPC at IDC. He may be reached at editor@ScientificComputing.com.
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||HIGH PERFORMANCE COMPUTING|
|Date:||Mar 1, 2011|
|Previous Article:||Genomics on the petascale: how HPC is transforming biological research.|
|Next Article:||BioStack-LS Tesla GPU compute cluster.|