Interview: Russian Supercomputer Outperforms All Others on Graph500
This week T-Platforms announced that Russia’s most powerful supercomputer at Moscow State University outperformed all competitors during the recent Graph500 benchmark tests. To learn more, I caught up with Anton Korzh, a systems architect at T-Platforms.
insideHPC: How is it that computer system from T-Platforms installed at the M.V. Lomonosov Moscow State University was able to achieve performance leadership in the Graph500 test?
Korzh: The results shown by the ”Lomonosov” system in the Graph500 test are primarily evidence of the versatility of the T-Platforms systems in various modes. This test is designed to identify the actual system performance when processing large amounts of data, where the outcome is influenced not only, and not even so much, by processor power, but by access time and data processing speed. Therefore, the only solutions able to show outstanding results in this test will optimally combine CPU speed, system interconnect, memory performance, and many other factors. An example of such a system is the “Lomonosov” supercomputer, the power of which, as a result of recent modernization, has been increased to a record-breaking for Russia 1.3Pflops.
insideHPC: Is T-Platforms solutions architecture optimized for handling of large volumes of data?
Korzh: Our solutions architecture is best suited to address a broad range of tasks, including, as shown by the Graph500 test results, processing of large data sets. For example, a lot of scientific groups are now solving different problems using the “Lomonosov” supercomputer complex: from the development of new nanomaterials and modeling protein molecule structure to analysis of global climate processes on the planet. Some of these tasks require a lot of computing power from processors and graphics cards, and others require processing of huge amounts of data, where the processor speed is secondary. In our work, we pay great attention not only to technology excellence, but also to the versatility of our solutions. Therefore, the systems from T-Platforms demonstrate consistently high performances in different tests.
insideHPC: Over the past year, we have seen leadership of Asian computing systems in the Top500 rating, but there was a Russian system that won the latest Graph500 test. Will, in your opinion, the trend of winning high positions in such benchmarks by Russian computer systems continue?
Korzh: If we analyze the Top500 ratings of recent years, the trend of strengthening Russia’s position in the global supercomputer industry becomes apparent. Our solutions are repeatedly included in the first hundred of the Top500 List, and the “Lomonosov” computer system, after a recent upgrade, has risen from the 17th to 13th position in the rating. Today, we are developing technologies that in the foreseeable future will be the basis of new, more powerful computing systems. So there is every reason to expect that domestic systems will continue to show ever-increasing performance, ranking higher and higher in the Top500 List. Leadership in the Top500 ranking is determined by the capacity of the system, which depends on the amount of funds invested in it. Thus, the Asian leadership in the Top500 List is determined only by the large amounts of funding allocated. By some estimates, creation of the Japanese K computer cost about one billion U.S. dollars. At the same time, the leadership in Graph500 can be achieved without such large investments – new original ideas, as well as new models and programming paradigms can give a huge effect, as we have seen in “Lomonosov’s” Graph500 results.
insideHPC: How much do you think the benchmark system used by the Graph500 creators influences the development of the supercomputer industry?
Korzh: This test conveniently illustrates the effectiveness of current systems for processing o data sets. It allows developers to optimize their solutions for such problems and increase their flexibility. In addition, the possible results of this test will contribute to the emergence of highly specialized systems focusing on high-speed data processing.
insideHPC: Big Data is said to be the next frontier of Cloud Computing and HPC. Do you agree with that? What is T-Platforms doing to optimize its position in this area going forward?
Korzh: Perhaps, this will really happen in the future, but so far we have not observed the obvious trends towards this. At present, HPC, Cloud Computing and Parallel Computing lines are developing independently, complementing each other in some areas.