GOVERNMENT
SKIF MSU becomes the most powerful supercomputing system in Russia
Moscow State University named after M. V. Lomonosov (MSU), Program Systems Institute of the Russian Academy of Sciences (PSI RAS), and T-Platforms company announce the completion of the SKIF MSU supercomputer project developed under the SKIF-GRID supercomputer program of Russia-Belarus Union. The SKIF MSU supercomputer with 60TFlops of peak performance has become the most powerful supercomputing system in Russia, CIS and Eastern Europe countries. Boris Gryzlov, Chairman of the State Duma (Parliament) of the Russian Federation, Chairman of Unified Russia party took part in the opening ceremony of the new MSU supercomputer. 
“Unified Russia party believes that the development of the Russian supercomputer industry, as well as building the network of supercomputing centers, is one of the key concepts underlying the innovative, knowledge-based economy of Russia. Without high performance computing today we are unable to develop either science-intensive technologies, nor competitive manufacturing, – said Boris Gryzlov. – As a result of successful achievement of the goals set by Russia-Belarus SKIF and SKIF-GRID supercomputer programs actively supported by the Party, today we are able to build supercomputers powerful enough to complete the most complicated computational tasks independently from the Western supercomputer vendors. We count that achievement as one of the most important components in the basement of the new Russian economy.” “The strategy of creating in-house supercomputer industry in Russia started to take shape, – said Andrey Kokoshin, Deputy Head of the Unified Russia fraction in Duma, member of RAS. – In the next few years it is urgent to have at least fifteen supercomputer centers based in leading Russian scientific and educational institutions, similar to that built in MSU with support from Unified Russia. Russian science and industry are able to solve the task of transforming Russia into one of the top 5 leading supercomputer states of the world, and we should set this task. The key components of success today are skilled strategic planning in this field, concentration of political power and proper management solutions.” The supercomputer SKIF MSU with 60TFlops of peak performance is based on 625 blades designed by T-Platforms and manufactured by Chenbro, incorporating 1250 quad-core Intel Xeon E5472 processors. The Linpack real performance test result is 47.17 TFlops or 78.6% of peak performance, which is the best efficiency among all quad-core Xeon-based systems in the first hundred of the Top500 global supercomputer list (www.top500.org). Actual performance of SKIF MSU corresponds to the 22nd position of the current Top500 list which is at the moment is a record result for Russia. According to current Top500 statistics, SKIF MSU is number 7 worldwide among supercomputers installed in the academic sector. Back in 1999 MSU became the first university in Russia to start building powerful supercomputing systems based on cluster technologies. “A wide range of pioneer scientific tasks is already being solved by MSU research teams which demand tens TFlops, – noted MSU dean and member of RAS Viktor Sadovnichiy. – Scientific teams of MSU and other institutions have already started employing the new machine for research in the fields of nanotechnology, fluid dynamics, quantum chemistry, seismology, drug design, climate, cryptography, and others.” “SKIF MSU is the most important result of the two years work on the SKIF-GRID program of Russia-Belarus Union. The technical solutions found form basis of the 4th generation of SKIF supercomputers, – said Sergey Abramov, Director of PSI RAS, member of RAS, scientific lead of SKIF and SKIF-GRID programs. – SKIF MSU will provide 60% of computing power of the SKIF Polygon distributed computing system, which by this summer will give over 100 TFlops of peak performance. This supercomputer will become a platform for SKIF-GRID RnD, as well as a key tool for research done by MSU and its partners all over Russia.” SKIF MSU supercomputer incorporated the most recent technological findings of the industry and uses several in-house developed technologies. Its computing core uses the first Russian-developed blade systems which incorporate 20 quad-core Intel Xeon 3.0GHz 45-nm processors in a 5U chassis, providing the highest computing density among all Intel-based blade solutions on the market. T-Blade is the first solution in the industry integrating new Intel 4500 chipset capable of providing up to 30% performance gain for real-life applications. In contrast to standard blades, T-Blade modules are compatible with any standard types of interconnect and other external network devices thanks to the PCI-Express 2.0 slot. SKIF MSU uses DDR InfiniBand with Mellanox 4th generation chips as a system network. Its architecture cuts down latency up to 1.2 usec and improves applications scalability, while providing compatibility with the new 40GB/sec QDR InfiniBand standard. This will save time, effort and money for the MSU team, as the applications developed and tuned for SKIF MSU will keep all performance gains when the university switches to a more efficient interconnect standard of the near future. SKIF MSU also applies Russian cluster software developed under the SKIF-GRID program. The package includes an in-house developed ready-to-use Linux distribution by ALT Linux, cluster and grid applications development tools by PSI RAS and MSU, and a robust management&monitoring system with provides easy access to all subsystems of SKIF MSU via an intuitive web interface. SKIF MSU is a fully integrated and balanced solution incorporating 60TB of T-Platforms ReadyStorage ActiveScale Cluster parallel storage, tape backup storage, fault-tolerant monitoring and management sub-system, complex power supply and cooling sub-systems with N+1 redundancy. The system is designed to ensure maximum efficiency of user applications performance. For example, the supercomputer uses several types of compute nodes with different memory volume and disk space to satisfy individual requirements of various application classes. Most of compute nodes are diskless, which improves the overall system fault tolerance. Diskless operating system launch simplifies cluster administration, allowing single-time OS image change for any system updates. The overall power consumption of the supercomputer is 520 kW (up to 720 kW in peak), and all this power is concentrated on the area of less than 100 square meters, while the room temperature stays below 20°C. For heat abstraction T-Platforms designed a modular cooling system with airproof “hot corridor” between racks with computing nodes. The solution guarantees abstraction of up to 30 kW of heat energy from each rack. The UPS system has N+1 redundancy of all components and in the case of emergency ensures the temperature stays at the set level for no less than 10 minutes. The new MSU supercomputer is built according to the schedule of the SKIF-GRID Russia-Belarus supercomputer program. The cost was divided between MSU and SKIF-GRID supercomputer program which is financed through the budget of the Russia-Belarus Union. For more information, please visit its Web site.
“Unified Russia party believes that the development of the Russian supercomputer industry, as well as building the network of supercomputing centers, is one of the key concepts underlying the innovative, knowledge-based economy of Russia. Without high performance computing today we are unable to develop either science-intensive technologies, nor competitive manufacturing, – said Boris Gryzlov. – As a result of successful achievement of the goals set by Russia-Belarus SKIF and SKIF-GRID supercomputer programs actively supported by the Party, today we are able to build supercomputers powerful enough to complete the most complicated computational tasks independently from the Western supercomputer vendors. We count that achievement as one of the most important components in the basement of the new Russian economy.” “The strategy of creating in-house supercomputer industry in Russia started to take shape, – said Andrey Kokoshin, Deputy Head of the Unified Russia fraction in Duma, member of RAS. – In the next few years it is urgent to have at least fifteen supercomputer centers based in leading Russian scientific and educational institutions, similar to that built in MSU with support from Unified Russia. Russian science and industry are able to solve the task of transforming Russia into one of the top 5 leading supercomputer states of the world, and we should set this task. The key components of success today are skilled strategic planning in this field, concentration of political power and proper management solutions.” The supercomputer SKIF MSU with 60TFlops of peak performance is based on 625 blades designed by T-Platforms and manufactured by Chenbro, incorporating 1250 quad-core Intel Xeon E5472 processors. The Linpack real performance test result is 47.17 TFlops or 78.6% of peak performance, which is the best efficiency among all quad-core Xeon-based systems in the first hundred of the Top500 global supercomputer list (www.top500.org). Actual performance of SKIF MSU corresponds to the 22nd position of the current Top500 list which is at the moment is a record result for Russia. According to current Top500 statistics, SKIF MSU is number 7 worldwide among supercomputers installed in the academic sector. Back in 1999 MSU became the first university in Russia to start building powerful supercomputing systems based on cluster technologies. “A wide range of pioneer scientific tasks is already being solved by MSU research teams which demand tens TFlops, – noted MSU dean and member of RAS Viktor Sadovnichiy. – Scientific teams of MSU and other institutions have already started employing the new machine for research in the fields of nanotechnology, fluid dynamics, quantum chemistry, seismology, drug design, climate, cryptography, and others.” “SKIF MSU is the most important result of the two years work on the SKIF-GRID program of Russia-Belarus Union. The technical solutions found form basis of the 4th generation of SKIF supercomputers, – said Sergey Abramov, Director of PSI RAS, member of RAS, scientific lead of SKIF and SKIF-GRID programs. – SKIF MSU will provide 60% of computing power of the SKIF Polygon distributed computing system, which by this summer will give over 100 TFlops of peak performance. This supercomputer will become a platform for SKIF-GRID RnD, as well as a key tool for research done by MSU and its partners all over Russia.” SKIF MSU supercomputer incorporated the most recent technological findings of the industry and uses several in-house developed technologies. Its computing core uses the first Russian-developed blade systems which incorporate 20 quad-core Intel Xeon 3.0GHz 45-nm processors in a 5U chassis, providing the highest computing density among all Intel-based blade solutions on the market. T-Blade is the first solution in the industry integrating new Intel 4500 chipset capable of providing up to 30% performance gain for real-life applications. In contrast to standard blades, T-Blade modules are compatible with any standard types of interconnect and other external network devices thanks to the PCI-Express 2.0 slot. SKIF MSU uses DDR InfiniBand with Mellanox 4th generation chips as a system network. Its architecture cuts down latency up to 1.2 usec and improves applications scalability, while providing compatibility with the new 40GB/sec QDR InfiniBand standard. This will save time, effort and money for the MSU team, as the applications developed and tuned for SKIF MSU will keep all performance gains when the university switches to a more efficient interconnect standard of the near future. SKIF MSU also applies Russian cluster software developed under the SKIF-GRID program. The package includes an in-house developed ready-to-use Linux distribution by ALT Linux, cluster and grid applications development tools by PSI RAS and MSU, and a robust management&monitoring system with provides easy access to all subsystems of SKIF MSU via an intuitive web interface. SKIF MSU is a fully integrated and balanced solution incorporating 60TB of T-Platforms ReadyStorage ActiveScale Cluster parallel storage, tape backup storage, fault-tolerant monitoring and management sub-system, complex power supply and cooling sub-systems with N+1 redundancy. The system is designed to ensure maximum efficiency of user applications performance. For example, the supercomputer uses several types of compute nodes with different memory volume and disk space to satisfy individual requirements of various application classes. Most of compute nodes are diskless, which improves the overall system fault tolerance. Diskless operating system launch simplifies cluster administration, allowing single-time OS image change for any system updates. The overall power consumption of the supercomputer is 520 kW (up to 720 kW in peak), and all this power is concentrated on the area of less than 100 square meters, while the room temperature stays below 20°C. For heat abstraction T-Platforms designed a modular cooling system with airproof “hot corridor” between racks with computing nodes. The solution guarantees abstraction of up to 30 kW of heat energy from each rack. The UPS system has N+1 redundancy of all components and in the case of emergency ensures the temperature stays at the set level for no less than 10 minutes. The new MSU supercomputer is built according to the schedule of the SKIF-GRID Russia-Belarus supercomputer program. The cost was divided between MSU and SKIF-GRID supercomputer program which is financed through the budget of the Russia-Belarus Union. For more information, please visit its Web site.