Exascale Plans for Russia

Print Friendly, PDF & Email
A Contributed Article by Alexey Komkov, Deputy General Director of Products and Technology, T-Platforms

The Concept of HPC development based on exascale-level supercomputer technology identifies the main evolution direction of the HPC industry in the years 2012-2020. The basis for this Concept is the result of the work performed by Rosatom State Corporation in cooperation with high-tech industries enterprises, as well as Russia’s leading scientific and educational centers. The Concept reflects the key proposals of major participants in the global HPC community concerning the need for and feasibility of creating computing systems of the next generation. Implementation of this concept should ensure technological breakthroughs in a number of strategically important sectors of the economy, including energy, nuclear physics, satellite navigation and communications, medicine and pharmaceuticals, as well as exploration.

Over the last 10-12 years, the performance of supercomputers has increased more than 1,000 times, and analysts say it can overcome the1 Exaflop barrier (1018 operations per second) already by 2018-2020. However, you can not ignore a number of constraints. They are mostly related to issues of systems’ power consumption, reliability and structural envelope. In this regard, exaflop-level computing clusters are expected to be built using hybrid architectures. All the work within the framework of the Concept is supposed to be performed in three phases including the development of supercomputers with a processing capacity of 10 Pflops and 100 Pflops by 2015 and 2017 respectively. Completion of the third phase, in which a system will be developed with a capacity of 1 Exaflop, is scheduled for 2020. The system will be built using processors with more than 100 cores, and its power consumption will be not less than 50MW. There is a plan within the Concept framework to develop a range of new technologies and high-tech solutions – including a new processor, a system of liquid cooling based on hot water flows, as well as system software and the environment for programming and cluster management. Dedicated application software will allow the new system to perform complex modeling of various processes on the basis of the latest techniques. In the process, our company, being one of the Concept developers, will create hardware and software systems and ensure their further development, including supercomputer monitoring and administration systems. Other partners involved in the Concept implementation will assume the development of programming environments.

In addition to development of the next-generation computer system, the Concept stipulates the progress of federal and regional Russian supercomputer centers to provide computing resources for enterprises and organizations in various industries. These centers will also contribute to the development of remote access technologies and expertise in supercomputing modeling, expand the range of services for applications, and provide training to prepare highly qualified personnel for the supercomputing industry. We plan to develop federal centers on the basis of the largest research centers in Russia: Computing Center of M.V. Lomonosov Moscow State University, Computing Center of the Russian Academy of Sciences, Computing Center of Kurchatov Institute Research Centre, and many others. These supercomputer centers will provide up to 100 Tflops in the first deployment phase, and up to 1 Pflops in the second one. We are planning first introductions of systems with a capacity of 1 Exaflop for the period of 2018-2020. In addition, there are plans to establish a network of 10-12 of Russia’s leading universities and academic and industrial organizations to implement a program to train skilled professionals in the field of exaflop technologies. On the basis of RAS institutions joined into the network, R&D works will be organized, and the faculty and research laboratories established for the intensive personnel training in the field of exaflop calculations. Creating of the university part of the network can be accomplished through the expansion of supercomputing technology centers established in 2010-2011.

Our company, along with leading players in the global HPC market, is a part of the Bureau of the Consortium forming the European technology platform designed to provide the EU with an independent access route to the strategic supercomputer technologies. A long-term plan for the hardware and software development generated by the Platform will be aimed at achieving a number of key objectives, such as leadership of Europe in scientific discoveries committed through the use of HPC technologies; high-accuracy projections for socially significant events; and creating energy-efficient technologies for the industry. The Platform will also help to design a new supercomputer component base that is based on the most competitive European developments while contributing to improved expertise in the use of HPC for industry and science purposes.

In addition, our company is a Russian representative in the Holistic Performance Analysis System (HOPSA) which has brought together the expertise of Russian and European supercomputing and scientific research centers for joint developments. The main objective of this project is to create an integrated infrastructure for comprehensive performance analysis and identification of bottlenecks in parallel computing implementations. Performance analysis at the individual processes level will allow for the development of best practices for improving both software and hardware system configurations. As a result, developers will have access to a rich toolkit to carry out in-depth performance analysis both of individual applications and the entire system. Russian HOPSA participants are responsible for the performance analysis and configuring of the computing system hardware, whereas their European colleagues have to perform a number of application optimization tasks. Thus, it will be possible to obtain reliable information both on ineffectiveness of individual applications, and on potential performance bottlenecks within system hardware units. As a result, a set of tools will be developed for supercomputer applications optimization and end-to-end performance monitoring for the entire system.

In addition to participating in these projects, we are developing our own next-generation supercomputer technologies. When creating new solutions, we rely primarily on our own development, thus preserving independence from foreign manufacturers of most components. On this basis, we plan to continue the development of international cooperation with developers of proprietary technologies contributing to next-generation computer systems.