Russia steps up game in supercomputing
Thanks to a 200-employee company based in Moscow, Russia is gaining attention in high-performance computingBy Patrick Thibodeau Computerworld - Russia's profile in supercomputing is rising, thanks to a Moscow-based company and a Russian president who sees high-performance computing as critical to the nation's future.
Two years ago, Russia's president, Dmitry Medvedev, somewhat sarcastically disparaged his country's lack of progress in the development of supercomputers, saying in a speech that "a huge number of entrepreneurs, not to mention officials, do not know what supercomputers are."
Medvedev is very interested in duplicating U.S. achievements in high technology. Last year, he traveled to Silicon Valley, where he visited Google, Twitter, Apple and Cisco and spoke at Stanford University about the need to increase technology investment in his country.
Medvedev has also visited Moscow State University to see a supercomputer built by T-Platforms, a 200-person, Moscow-based high-performance computing (HPC) company. The T-Platforms system at Moscow State University is the 13th most powerful supercomputer in the world, as ranked in the most recent Top500 list.
The T-Platforms system also placed third on a new benchmark, the Graph500, which measures how rapidly a computer can execute a data-intensive graph operation. The Top500 list, which uses the Linpack benchmark, measures how fast a computer solves a series of dense linear equations.
Steve Conway, an HPC analyst at research firm IDC, said Medvedev has been calling for investments in Russian high-performance computing systems. The president is concerned that, without such investments, "Russian products in five years won't be competitive in the world market," said Conway. "So the government is very much driving to increase HPC usage."
T-Platforms is promoting supercomputing use in Russia and is selling its systems in Europe. It hopes to break into the U.S. market someday.
"At the moment, it's a new technology for most of the people, but we are trying to expand the market [in Russia]," said Anton Korzh, a systems architect at T-Platforms. The company is also helping its potential science and research customers to design applications that work in HPC environments.
The Graph500 ranking was developed by an international team led by the U.S. Department of Energy's Sandia National Laboratories, and it made its first appearance last year. There were only 28 systems ranked on last month's latest list, and although T-Platforms' 1.3-petaflop system, called Lomonosov, placed third, it had the highest performance. Its overall ranking was hurt by the size of the problem it was tested on.
An IBM system -- its Blue Gene/P, a.k.a. the Intrepid -- placed first on the Graph500.
Korzh said he views the Linpack Top500 test as directly related to the amount of money spent to build a system. "The Linpack performance can almost be predicted," Korzh said. But with the Graph500 it's a different story, because software technologies have more of role in determining performance. "You have more possibilities of optimizations," he said.
Conway said software capability is extremely important. "HPC leadership is going to be determined going forward much more by software advances than by hardware advances -- the hardware has already raced way out of the ability of software to exploit it," he said.
When it comes to writing software, "all of Eastern Europe has some advantage, because all through the time they were under communism they had very poor hardware, so they had to write software that would practically make a washing machine compute," Conway said.
Conway said he's not saying that Russian and Eastern European developers will "outrun or outgun those in the Europe or the U.S.," but "they definitely have the advantage of developing software that is robust enough that it can withstand a lot."
Korzh said that the Graph500 was run on only half the nodes of T-Platforms' hybrid blade system, which runs Nvidia Tesla GPUs. The company plans to try again on a larger set of nodes for the November Graph500 ranking.