Translated by

RSK Tornado

Product
Developers: RSK Technologies
Last Release Date: 2019/11/28
Technology: Supercomputer

Content

The solution based on architecture of RSK Tornado is distinguished by high compactness (all cluster is located in two rack cabinets), reliability, scalability and energy efficiency. The indicator of PUE (the relation of the general energy consumption to energy consumption of the IT equipment) – less than 1.2, it means that no more than 17% of the consumed electric power are spent for cooling. Computing efficiency of a cluster in the LINPACK test – more than 90%.

2019: Computing RSC Tornado AFS RSC Tornado AP and DWH node

On November 29, 2019 the RSK company provided development of a line of the unified solutions RSC Tornado for accomplishment of a broad spectrum of resource-intensive scientific and applied tasks. The updated line of the integrated software-defined and reconfigurable solutions is focused on application as as a part of the classical high-performance systems (high performance computing, HPC), and for effective storage and data processing and also for creation of the systems of artificial intelligence (Artificial Intelligence, AI), the systems of machine and deep learning (Machine Learning, Deep Learning is ML/DL).

Provided to RSK development of a line of the unified solutions RSC Tornado

Among the solutions presented by specialists of RSK:

  • The computing RSC Tornado AP node based on high-performance server processors of a line Intel Xeon Platinum 9200 (to 56 cores on the processor).
  • The high-performance storage systems of RSC Tornado AFS intended for application both in the field of high-performance computing, and in areas of machine and deep learning. They are based on the advanced program stack of DAOS for creation of the distributed object storage systems with support of modules of a non-volatile memory of Intel Optane DC Persistent Memory.

According to the developer, development of a line of RSC Tornado allows to implement at higher level the main opportunities of solutions of RSK, for example, such as: the maximum computing density and energy efficiency (due to 100% of liquid cooling with "hot water" of all electronic a component), linear scalability from the small systems in several servers to thousands of servers as a part of big clusters or server farms. At the same time additional opportunities for optimization of cost of final solutions due to support of open standards including drives for storage systems are given:

and also:

  • server payments with support of bigger amount of RAM,
  • processors with the maximum energy consumption to 500 W on a socket,
  • a broad set of accelerators with energy consumption to 700 W.

As a result, according to the producer, the updated line of RSC Tornado will allow to create systems with a bigger bulk computing density, a wide variety of sets of the applied a component and their configurations for achievement of high efficiency of the specific solution. In turn, unification of a form factor of a mounting cabinet, including distributed system of power supply with duplication of N+x, the built-in system of monitoring and management computing and switching components, allows to use at the same time in one rack as solutions of RSK from 100% liquid cooling, and the server and communication hardware of a standard form factor 19' (rack unit, RU) other producers, equipped with the air or combined cooling.

Computing RSC Tornado AP node

According to the statement of RSK, the high-performance RSC Tornado AP node with support of 56 nuclear processors of a line of Intel Xeon Platinum 9200 (Intel Xeon Platinum 9282 model) and direct liquid cooling in the "hot water" mode has maximum theoretical (peak) with an efficiency of 9.3 TFLOPS, having at the same time 24 channels of RAM and providing support of volume of storage to 1.5 TB data. Such node can be completed with two solid state drives (SSD) with support of NVMe technology in a form factor of M.2 — for example, Intel Optane SSD DC P4801X M.2 Series or Intel of SSD DC P4511 (NVMe, M.2), or two SSD-drives of the E1.S format (short ruler) — for example, Intel of SSD DC P4511 (NVMe, E1.S). As an option expansion of a system using an additional basket with 6 SSD drives based on NVMe in forms - a factor of E1.L (long ruler) up to 15.36 TB everyone with a possibility of hot swap is possible.

So, for example, the configuration with Intel of SSD DC P4320/P4520 (NVMe, E1.L) gives the chance to provide the volume of storage of more than 100 TB data on a node with quick access. The optimal combination computing, network and components of data storage provides necessary balance for creation of the high-performance hyper convergent systems with linear scaling, as for achievement of necessary computing power, and required parameters of volume/speed of distributed system of storage. This approach allows to create the high-performance and compact systems having indicators, record for the industry: 0.8 PFLOPS gross peak capacities and 8.4 PB of volume of data storage in one mounting cabinet 42U, claim in RSK.

Storage systems of RSC Tornado AFS

Considering constantly growing customer needs in increase in volumes of storage and data processing rates, RSK developed the solution RSC Tornado AFS for creation of All-Flash of storage systems of large volume on the basis of high-speed NVMe technology and with the most dense form factor of EDSFF.L. The array provided to All-Flash from 100% liquid cooling of all a component with "hot water" supports to 32 solid state drives with support NVMe technologies in forms - EDSFF.L factor from stated for November, 2019 with a capacity of 15.36 TB everyone and a possibility of hot swap. The doubling of capacity of drives of the class NVMe/EDSFF.L expected in the near future will allow to increase storage volume up to 1 PB by 1 place of a standard form factor 19' (1RU) without any constructive changes, consider in RSK.

According to the producer, wide use of NVMe-over-Fabric (NVMeOF) technology gives opportunities for creation of high-speed distributed systems at data transfer rates to several TB / with and with a capacity of storage up to 20.64 PB on a cabinet with assistance of different types of parallel file systems, such as Lustre, BGFS, etc. Use of perspective Intel Optane DC Persistent Memory technologies and remote direct memory access of RDMA (remote direct memory access) opens a different approach for creation of the high-speed low-latent distributed data storage systems (DSS) of the class key-value store by means of the advanced program stack of DAOS (Distributed Asynchronous Object Storage). Such DWH is intended for wide use in areas of machine and deep learning.

For the purpose of achievement of optimal performance special attention was paid also to a basic part of the solution in which two high-performance processors Intel Xeon Scalable of the second generation are used the possibility of use up to 2 TB high-speed memories and to four modules Intel Optane DC Persistent Memory as caches of these 4-5 levels is implemented. The communication subsystem consisting of 2 PCIe Gen3/4 x16 adapters on the basis of technologies Intel Omni-Path, InfiniBand or Ethernet that provides access rate to data at the level up to 25 GB / with on an array is responsible for ensuring high-speed internodal exchange at speeds up to 200 GB / with, noted in RSK.

According to the statement of the producer, adding of modules of a non-volatile memory of Intel Optane DC Persistent Memory became logical development of a high-speed hyper convergent storage system of RSC Tornado HS about the 12th NVMe-by drives that gave sales opportunity of support of a program stack of DAOS.

Use of the hyper convergent solution RSC Tornado with a program stack of RSC BasIS allows to define by the built-in orkestrator system architecture of data storage "on the fly" after installation of the equipment, adapting at the same time a complex to different types of loadings according to preferences and tasks of users. At the same time there is possible a creation to "storages-on-requirement" (storage-on-demand) with different characteristics for each of them (volume, type of the file system, access rate, level of reliability and security, lifetime), claim in RSK.

2017

Training of RSK for emergence of the Purley platform

The RSK company developed the updated superdense, scalable and energy efficient cluster solution of RSK Tornado, it was provided on June 19, 2017 at the international conference ISC'17 in Frankfurt am Main. The solution represents set a component for creation of modern computing systems of different scale from 100% liquid cooling in the "hot water" mode. It includes high-performance computing nodes based on processors Intel Xeon Phi 7290 and Intel Xeon E5- 2697A v4 in combination with first-ever high-speed switch Intel Omni-Path with similar cooling on "on hot water".

Administration and monitoring of subsystems of RSK Tornado provides functionality of the integrated program stack "RSK BASIS" for management of cluster systems.

"RSK Tornado" on server Intel processors has high rates of compactness and computing density (up to 153 nodes in one standard cabinet of 80 cm x 80 cm x 42U), the high level of energy efficiency, provides a possibility of stable work of computing nodes in the "hot water" mode at a temperature of coolant up to +65 °C on an input in computing nodes and switches.

Work in the "hot water" mode in this solution allows to apply the year-round free cooling (24x365) mode, using only dry water-cooling towers working at an ambient air temperature up to +50 °C that[1] allows to get rid completely of a freon circuit and chillers[1].

Questions of TAdviser of RSK Tornado were answered by Alexey Shmelyov, the chief executive of RSK group.

Performance of RSK Tornado reached 685.44 TFLOPS

On July 18, 2017 the RSK group provided the superdense, scalable and energy efficient cluster solution to RSK Tornado on the basis of processors of the Intel Xeon Scalable family.

On the Intel Xeon processors "RSK Tornado" set a world record of performance for high-performance solutions — 685.44 TFLOPS in a standard computing cabinet 42U (80x80x200 cm). This indicator by 2.65 times exceeds performance of RSK Tornado on the platform of the most senior model of the processor of the previous generation of the Intel Xeon E5-2600 v4 family.

The specified performance is reached on the Intel Xeon processor Platinum 8180 (28 cores, the clock rate of a core of 2.5 GHz, the maximum energy consumption of 205 W, volume a cache memory of L3 level makes 38.5 MB) from the Intel Xeon Scalable family.

"RSK Tornado" is cooled with liquid

On June 19, 2017 the RSK group presented a cluster solution of RSK Tornado with direct liquid cooling - all elements of a computing cabinet, including switches of network, are cooled with liquid coolant.

To it the solution of RSK on the basis of the 72-core Intel Xeon processor Phi 7290 is characterized by computing density for architecture of x86 in 1.41 PFLOPS on a cabinet or more than 490 TFLOPS per cubic meter[2].

As the company stated, the next generation of RSK Tornado is ready to support server processors of the Intel Xeon Processor Scalable Family family (the code name Skylake-SP).

"RSK Tornado" on the basis of server Intel processors provides computing density up to 153 nodes in one standard cabinet of 80 cm x 80 cm x 42U), the energy efficiency, provides a possibility of stable work of computing nodes in the "hot water" mode at a temperature of coolant up to +65 °C on an input in computing nodes and switches.

2016

"RSK Tornado" reached computing density 1.41 PFLOPS

On November 16, 2016 the RSK group presented the supercomputer solution "RSK Tornado" with direct liquid cooling on the platform 72 the nuclear Intel Xeon processor Phi 7290. Development set a world record in 1.41 PFLOPS computing density on a cabinet for architecture of x86.

As a part of a configuration of the computing system:

Server payments, (2016)
Server payments, (2016)

Features of architecture of RSK Tornado

  • use of multi-core processors of the Intel Xeon Phi 7200 family, including the senior Intel Xeon Phi 7290 models (up to 72 cores) and a possibility of use of processors Intel Xeon Phi 7250F, Intel Xeon Phi 7290F (F suffix for versions of processors with the integrated high-speed Intel OmniPath internodal connection),
  • use of server payments of the Intel Server Board S7200AP family,
  • physical density with placement to 408 computing nodes in a bilateral cabinet 42U the sizes of 120х120х200 cm,
  • computing density is 1.41 PFLOPS (earlier 528 TFLOPS) in a bilateral computing cabinet 42U or more 490 TFLOPS/m ³,
  • power density up to 200 kW / a cabinet, due to decrease in energy consumption of a system helped to increase energy efficiency almost three times,
  • honor increase in amount of RAM of a decisive field of computing nodes of one cabinet by 5 times with 16.3 TB in the previous generation up to 76.5 TB (to 192 GB of RAM of the DDR4-2400 RAM type and 16 GB of MCDRAM on a node),
  • simultaneous use to two solid state drives with connection on the bus SATA and one PCIe in a form factor of M.2, as Intel of SSD DC S3500 and Intel of SSD DC P3100 (the M.2 NVMe),
  • the increased energy efficiency level – conditions for stable work of computing nodes in the "hot water" mode at a temperature of +63 °C on an input in computing nodes are provided that helps to ensure system operation in the mode of "freecooling" 24x365 about PUE systems less than 1.05,
  • power supply module in a form factor of a computing node, provides effective conversion of 220 V of alternating current to 400 V of a direct current (efficiency of conversion of a power supply of 96%) and a possibility of parallel operation on the common bus with reservation from N+1 to N+N,
  • the upgraded construction of a computing cabinet with support of technologies of high-speed internodal connections, including Intel Omni-Path and Mellanox EDR InfiniBand,
  • the possibility of creation of flexible configurations of cooling systems, with a possibility of reservation, both separate nodes of hydroregulation, and all system in general is provided,
  • any node of the solution "RSK Tornado" can individually be serviced and does not require a stop of other node. Convenient access to all components of a node (to memory, disks, adapters of high-speed interconnections, management subsystems and power supply) allows to make easily replacement of these components or their reconfiguration on the platform of the customer.

The cluster solution of RSK Tornado can be implemented on the basis of server processors of the Intel Xeon E5-2600 family, including the senior Intel Xeon E5-2699A v4 model (22 cores, clock rate of 2.40 GHz, 55 MB L3 cache memory), providing high computing density – 258.5 TFLOPS in standard computing to cabinet 42U (80x80x200 cm).

Density of calculations is lifted up to 528 TFLOPS

On June 21, 2016 the RSK company announced upgrade of the solution "RSK Tornado" using the Intel Xeon processor Phi.

The computing density of the solution grew twice to 528 TFLOPS on a cabinet. The upgraded solution of RSK has the improved indicators of physical and computing density, the high level of energy efficiency and ensures stable functioning in the "hot water" mode at a temperature of coolant of +63 °C.

Computing module \"RSK Tornado\", (2016)
Computing module "RSK Tornado", (2016)


Indicators of the operating system:

  • use of the senior models of the latest multinuclear (to 72 cores) Intel Intel Xeon Phi 7250, Xeon Phi 7290 or Intel Xeon Phi 7250F, Intel Xeon PhiT 7290F processors (F suffix for versions of processors with the integrated high-speed Intel Omni-Path internodal connection),
  • use of new server payments of the Intel Server Board S7200AP family,
  • the highest physical density with placement to 153 computing nodes in a standard cabinet 42U with sizes of 80х80х200 cm,
  • the computing density increased almost twice – 528 TFLOPS (earlier 280 TFLOPS) in a standard computing cabinet 42U or more 412 TFLOPS/m3
  • up to 192 GB of RAM on a node (DDR4-2400 RAM + 16 GB of MCDRAM),
  • simultaneous use to two solid state drives with connection on the bus SATA and one PCIe in a form factor of M.2, such as Intel of SSD DC S3500 and Intel of SSD DC NVMe M.2,
  • the increased reliability – independent hydraulic pumping modules (hydroregulation modules) of a liquid cooling system on each computing domain (up to only 9 modules on a cabinet) with reservation from N+1 to N+N,
  • the increased energy efficiency level – necessary conditions for stable work of computing nodes in the "hot water" mode at a temperature of +63 °C on an input in computing nodes are provided,
  • the new module of power supply in a form factor of a computing node providing highly effective conversion of 220 V of alternating current to 400 V of a direct current and a possibility of parallel operation on the common bus
  • the updated construction of a computing cabinet with support of new technologies of high-speed internodal connections, including Intel Omni-Path and Mellanox EDR InfiniBand,
  • the possibility of creation of flexible configurations of cooling systems, with a possibility of reservation of separate nodes of hydroregulation and all system in general is provided.

2015: The cluster of RSK Tornado is provided

On July 13, 2015 the RSK group provided the next generation of the cluster solution of RSK Tornado.

"RSK Tornado" has the improved indicators of compactness and computing density, energy efficiency.

"RSK Tornado", 2015


Solutions on the platform of the company of a cluster architecture RSK Tornado developed by specialists, with liquid cooling, are in operation at the Russian customers more than four years. They are set in Peter the Great St.Petersburg Polytechnic University (SPbPU), the Interdepartmental supercomputer center of the Russian Academy of Sciences (JSCC RAS), the Southern Ural State University (SUSU), the Moscow physics and technology university (MIPT), Roshydromet, at other customers from different industries.

The cluster solution of RSK Tornado has characteristics:

  • physical density – is increased to 153 computing nodes on a cabinet
  • increase in computing density – more than 200 TFLOPS/m3 on standard processors and up to 256 GB of RAM on a node,
  • reliability augmentation – independent hydraulic pumping modules (hydroregulation modules) of a liquid cooling system on each computing domain (up to only 9 modules on a cabinet) with reservation from N+1
  • increase in level of energy efficiency – necessary conditions for stable work of computing nodes in the "hot water" mode at a temperature of +65 ° C at the exit from a node (that is today a world record in the NRS-industry) are provided,
  • power supply module in a form factor of a computing node, provides highly effective conversion of 220 V of alternating current to 400 V of a direct current and a possibility of parallel operation on the common bus,
  • construction of a computing cabinet with support of new technologies of high-speed internodal connections, including Mellanox EDR Infiniband, Intel Omni-Path is updated,

The solution assumes support of future Intel Xeon processors and Intel Xeon Phi with code names of architecture Broadwell and Knights Landing.

High availability and fault tolerance are provided at the expense of a management system and monitoring of work, both separate nodes, and cluster system in general, enhanced capabilities on management of power supply, ensuring reservation of power supplies and modules of hydroregulation. All elements of a complex (computing nodes, power supplies, hydroregulation modules, etc.) have the selected managing controller that provides ample opportunities for telemetry and management of each element.

The construct of a cabinet allows "hot swap" of modules of hydroregulation without interruption operability of a complex. Liquid cooling of all components provides long term of their service.

The advanced technology approaches implemented in new generation of a cluster solution of RSK Tornado allowed to reduce infrastructure cost within project implementation of creation of computer systems and to provide opportunities for more flexible upgrade as at the level of a separate node, and all system.

The new generation of RSK Tornado is constructed on the basis of conventional server components of Intel – the server Intel Xeon processors E5-2600 v3, server payments of Intel S2600KP and solid state drives of Intel of SSD DC S3500/3600/3700 for data processing centers.

According to the statement of the company, the cluster solution of RSK Tornado continues to be in the lead in the industry on indicators of physical and computing density, energy efficiency, reliability, availability and controllability.

"Unique long-term experience of specialists of RSK in development of technologies of highly effective direct liquid cooling and superdense integration of supercomputer solutions based on conventional server components allowed to develop and provide new generation of a cluster solution of RSK Tornado with a number of the improved characteristics which are very demanded by the customers operating powerful computer centers. In addition to the world records of computing and power density which are earlier set by our solutions on the occupied volume, on new generation of RSK Tornado the world record of stable work in the "hot water" mode at a temperature is set +65o by Page. All developments of RSK become in Russia, in production of our products we actively rely on the potential and production capacities of the Russian industrial enterprises", – Alexey Shmelyov, the chief executive of RSK group noted.

2013: Development of architecture

The new round of development of architecture of RSK Tornado for creation of the energy efficient and compact data processing centers (DPC) and supercomputer complexes allowed specialists of RSK group for the first time in the world to implement direct liquid cooling for standard and in large quantities available server payments (different producers) on the platform of the Intel Xeon processors which are initially created for the traditional systems with air blowing electronic a component together with the latest Intel Xeon Phi coprocessors, the press service of the company-razraotchika reported on July 4, 2013.

This third generation of energy efficient solutions of RSK for segments of high-performance and cloud computing and also DPC.

High-performance solutions with a high computing density on the basis of architecture of RSK Tornado with liquid cooling are intended for the solution of different tasks of customers.

The product line includes:

  • RSK of microdPC (from 16 to 64 nodes),
  • RSK of minidPC (from 64 to 256 nodes),
  • RSK of DPC (more than 2 racks with high density up to tens petaFLOPS).

Characteristics

  • up to 128 h86-servers in a standard 42U rack of 80х80х200 cm;

  • high density design of blade servers on the basis of standard and in large quantities available server payments;

  • the highest energy efficiency — a performance indicator of use of the electric power of PUE (Power Usage Effectiveness) reaches value 1.06, record for the HPC-industry (a ratio "energy consumption of all system / energy consumption electronic a component"). So no more than 5.7% of energy consumption are spent for cooling of all system;

  • the record coefficient of computing efficiency reaches 96% on the LINPACK test for the new Intel Xeon processors E5-2690 (the Intel Turbo Boost technology works all the time that provides a gain of clock rate to 400 MHz during the work with the LINPACK test);

  • removal more than 100 kW of thermal power from a rack with the help of a unique liquid cooling system of RSK;

  • possibility of use of the most high-performance models of server Intel processors with heat release of 135 W. For example, Intel Xeon processor E5-2690 (2.9 GHz, 8 cores) and the latest high-performance Intel Xeon Phi 7120X and 5120D coprocessor (1.23 GHz, 61 cores);

  • high peak computing power is more than 47 TFLOPS in one rack based on architecture of Intel x86 with a set of instructions of Intel of AVX and more than 200 TFLOPS using Intel Xeon Phi coprocessors;

  • high density — 74 TFLOPS on sq.m. (only based on Intel Xeon processors) and 312 TFLOPS on sq.m. (with Intel Xeon Phi coprocessors);

  • high scalability – to the level of several petaFLOPS (tens of racks);

  • cost efficiency – reduction of operating costs up to 60% (economy of costs for the electric power in rub thanks to operation of the solution of RSK);

  • compactness – reduction of the area of DPC several times in comparison with traditional solutions on the basis of air cooling;

  • possibility of use of accelerators and coprocessors (for example, Intel Xeon Phi).

  • the complete integrated software stack "RSK BASIS", optimized for high-performance computing.

  • performance and scalability of solutions on the basis of architecture of RSK Tornado is confirmed with the certificate of Intel Cluster Ready.

Notes

212