Customers: Interdepartmental supercomputer center RAS (JSCC RAS) Contractors: RSK (group, earlier - RSK the Scythian), RSK Technologies Product: MVS-10P OF MEGAPIXELНа базе: RSC PetaStream Project date: 2017/01 - 2017/05
Project's budget: 144 048 000 руб.
|
2017
upgrade of SK Is complete
On April 26, 2017 the RSK Technologies company announced end the project of upgrade of computing resources of the Interdepartmental supercomputer center of Rossiiysky academy of Sciences (JSCC RAS).
In JSCC RAS universal computing cabinets of "RSK of the Tornado" of high power density and the system of precision liquid cooling are set. In accordance with the terms placements of the equipment for JSCC RAS the optimal operation mode of a computing cabinet is selected at a continuous temperature of coolant of +45 °C on an input in computing nodes (with peak value to +57 °C). Work in the "hot water" mode in this project allowed to apply the year-round free cooling (24x7x365) mode, using only dry water-cooling towers working at an ambient air temperature up to +50 °C and also to get rid of a freon circuit and chillers. As a result the annual average indicator of a PUE system reflecting the level of efficiency of use of the electric power makes less than 1.06 - less than 6% of the consumed electricity[1] are spent for cooling[1].
At the heart of computing nodes the powerful 72-core server Intel Xeon processors Phi 7290, the 16-core server Intel Xeon processors E5-2697A v4, server payments of the Intel Server Board S7200AP and Intel Server Board S2600KP families, solid state drives of the SSD DC S3500 Intel family with connection on the bus SATA in a form factor of M.2.
For high-speed data transmission between computing nodes as a part of a supercomputer complex JSCC RAS the technology of high-speed switching Intel Omni-Path providing the speed of not blocked switching to 100 Gbit / c on the basis of the 48-port Intel Omni-Path Edge Switch 100 Series switches is used. It will allow specialists to satisfy JSCC RAS of requirement of resource-intensive applications of users, to provide a necessary stock of network transmission capacity in the long term.
The Intel Omni-Path Architecture technology (Intel of OPA) represents the complete solution for high-speed switching and data transmission designed to help to increase with the minimum costs performance of operation of applications in HPC-clusters of initial level and large-scale supercomputer projects. the 48-port switch of Intel of OPA allows to connect at the identical budget 26% more servers in comparison with the competing solutions and also to reduce energy consumption to 60%, providing, thus, more energy efficient switching and system infrastructure.
Availability, fault tolerance and ease of use of the computing systems created on the basis of technologies of RSK for high-performance computing are provided by means of a management system and monitoring on the software platform "RSK BASIS". It allows to manage separate nodes and all complex, including infrastructure components.
All elements of a complex (computing nodes, power supplies, hydroregulation modules, etc.) have the built-in module of management that provides ample opportunities for detailed telemetry and flexible management. The construct of a cabinet allows to replace computing nodes, power supplies and hydroregulations (at the condition of application of reservation) in the mode of hot swap without interruption of operability of a complex. The majority of components of a system (such as computing nodes, power supplies, network and infrastructure components, etc.) is the software-defined components allowing to simplify and accelerate installation and service, the subsequent upgrade of a system. Liquid cooling of all components provides long term of service.
Upgrade of SK
On January 31, 2017 it became known of the forthcoming project of upgrade of power of the main calculator in JSCC RAS.
Accumulation of computing power is carried out within contract performance based on the tender of December, 2016 for delivery of the corresponding equipment. Works in the project are performed by RSK Technologies company. The agreement is signed without decrease in the starting price — 114.48 million rubles.
MVS-10P MP – Expansion of a complex JSCC RAS on the basis of the massive and parallel solution RSC PetaStream, (2015)
Till March 31, 2017 RSK should be delivered in MSTs not less than 88 computing nodes of gross peak capacity not less than 102 Tflops[2].
At the same time in December, 2016 RSK Technologies became the only participant of an auction on conducting major repair of MVS-10P. Within the contract received according to the results of the tender for 30 million rubles, the companies should be replaced till March 31, 2017 27 computing nodes which inadequate work (as it appears from the technical project), has negative effect on the actual output of the operating (not upgraded) system.
2015: SK expansion
On April 15, 2015 it became known of the project of expansion of a supercomputer complex in JSCC RAS the solution MVS-10P MT of RSK[3].
Building JSCC RAS, 2014
Project Results
The RSK group conducting development and integration of solutions for the data processing centers (DPC) announced installation of the powerful MT MVS-10P computing system in the Interdepartmental supercomputer center of the Russian Academy of Sciences (JSCC RAS).
MVS-10P MP - the supercomputer with a peak capacity of 77.33 TFLOPS took the 20th place in the new edition of the rating of Top50 of the most powerful computing systems in Russia and the CIS. The MT MVS-10P system became expansion of the supercomputer complex (which basis the MVS-10P supercomputer on the platform of a cluster architecture of "RSK of the Tornado") created by RSK in 2012 (takes the 4th place in rating). After project completion of installation and start, the gross performance of the supercomputers of RSK with liquid cooling installed in JSCC RAS - more than 600 TFLOPS.
MT MVS-10P supercomputer, 2014
Architecture of RSC PetaStream on the basis of which MVS-10P MP, completely Russian development is created. Feature of RSC PetaStream - implementation of the concept of the massive parallelism providing accomplishment up to 250 thousand parallel performed flows on 1024 computing nodes of architecture of x86 in one computing cabinet on the area of 1 sq.m. RSC PetaStream possesses world records on computing – 1.2 petaFLOPS on a cabinet – and power – more than 400 kW on a cabinet – density, on performance on the occupied volume – more than 560 TFLOPS/m3 and compactness.
The server products of Intel used during creation MVS-10P MP: 61-core Intel Xeon Phi 7120D coprocessors, processors of the Intel Xeon E5-2600 family, server payments of Intel S1600JP, solid state drives of Intel of SSD DC S3500 for corporate DPCs. The total amount of memory (GDDR5) on nodes - 1 TB, internodal connection - 56 Gbit / c.