(H)igh (P)roductivity (C)omputing. HPCwire

Cluster Computing:

Multi-Teraflop Cluster to be Deployed in Russia

Intel Corporation, Microsoft Corporation and T-Platforms Company, a Russian developer and manufacturer of integrated solutions for high performance computing and data centers, announced the launch of the project for building a supercomputer, codenamed SKIF Cyberia, based on dual-core Intel Xeon processors. The system is unique in the CIS (Commonwealth of Independent States) and Eastern Europe countries in terms of technologies, performance and design and will serve as the basis of a regional center for high performance computing resources at Tomsk State University (TSU) in Russia.

The basis for the T-Platforms SKIF Cyberia supercomputer system will be a 283-node high performance cluster based on 566 dual-core Intel Xeon 5150 processors. The system will attain a peak performance of the 12 TFlops, with power consumption being 90 kW, at most. The T-Platforms supercomputer for TSU will be the first high performance system in CIS based on the new dual-core Intel Xeon processors as well as one of the first systems of that kind in the world. The Linpack test used to evaluate actual predicted performance of SKIF Cyberia, showed a result of 7.8 TFlops, which will enable TSU system to become not only the most powerful supercomputer within Russia, CIS and Eastern Europe countries, but also one of the most powerful computers in the world.

The delivery set will also include SUSE Linux Enterprise Server 9 Operating System, freeware control and monitoring tools, Fortran and C/C++ Intel compilers, Intel VTune Performance Analyzer 8.0 optimization tools and Intel Math Kernel Library 8.1.

Intel started shipping its dual-core Xeon 5100 processor based on a new Intel Core architecture in May of 2006, while the industry's first four-core Intel Xeon 5300 processor started shipping in November. In late October of 2006 in San-Francisco Intel made a trial run of the system based on the four-core processor, codenamed Tigerton, for multiprocessor server systems, which will finalize the switching of Intel Xeon processor family to the power-saving Intel Core microarchitecture in all server market segments. The Tigerton processor, targeted for 3Q 2007, is a 65-nm processor used for multi-processor servers within a new Caneland platform.

SKIF Cyberia will be the first supercomputer in CIS and one of the first clusters in the world to use the new Microsoft Windows Compute Cluster Server 2003 operating system, designed specifically for concurrent computing. An innovative approach of Microsoft enables Windows Compute Cluster Server 2003 to run under the Linux OS as well as make it possible to integrate the cluster with existing Windows OS-based infrastructure at TSU.

"Windows Compute Cluster Server 2003 allows for high performance computing to be made at any level -- starting from personal systems and lab- or department-level systems up to large-scale supercomputer systems for multiple-access centers," said Kiril Fayenov, Director of HPC program at Microsoft Corp. and manager for development of Microsoft Windows Compute Cluster Server 2003. "Indeed, of prime importance for us is the interaction with developers and resellers of large-scale systems which helps to solve engineering problems encountered in designing world-class high-performance systems and demonstrate Windows capabilities on powerful computing clusters. That's why we are pleased to co-operate with T-Platforms Company which has a unique background in Russia in designing large-scale systems within the framework of SKIF program."

The TSU supercomputer is the first in CIS and Eastern Europe to use a QLogic InfiniPath technology with low latency (as low as 1.3 us) at a throughput of up to 950 Mb/sec. Cluster control will be carried out using Gigabit Ethernet and SKIF ServNet networks. Cluster computing node RAM will total 1.1 TB, with an internal disk capacity of 22.5 TB. For external storage, SKIF Cyberia is using NAS storage with T-Platforms ReadyStorage Active Scale Cluster parallel file system of 10 TB. The T-Platforms supercomputer will be provided with infrastructure solutions unique for Russia and CIS including an integrated 120 kW UPS and conditioning system using up-to-date water cooling technologies with 96 kW capacity.

As one of the winners of all-Russian contest for innovative educational programs held within the framework of the National Education project, TSU had the opportunity to create a regional center for multiple access to high performance resources. The bid of T-Platforms Company for a state contract for delivery of hardware to implement the above project was won in an open contest.

"The regional center of multiple access to high performance resources under TSU based on T-Platforms SKIF Cyberia supercomputer will be a key resource in implementing Tomsk Region development strategy," reported Georgy V. Mayer, rector of Tomsk State University, D.SC., phys. and math., professor. "One of the guidelines of this strategy is to increase the extraction of minerals: As of today only a half of proven oil and gas fields are in commercial operation while potential hydrocarbon resources in the region exceed the current production by three times. Other important challenges include maintaining competitiveness of products manufactured in instrument-making and machine-building enterprises of the region, which have not long ago been mainly oriented to the needs of military-industrial complex, as well as enterprises of petrochemical and pharmaceutical industry; ecological monitoring and rational use of forest resources, deposits of mineral therapeutic water and artesian springs, climatic and weather research. The TSU center of high performance computing will also provide for a competitive level of research in the fields customary for Tomsk such as biotechnology, plasma and high-energy physics, modeling of new materials, mesomechanics and much more."

The state contract for delivery of hardware to TSU has been signed by T-Platforms Company in September 2006; the bid price was 50967 thousand rubles. Thus, the price/peak performance ratio of the TSU supercomputer will be as little as $158 thousand per 1 Tflops which is 4.5 times less than that of 288-node cluster, SKIF K-1000, supplied by T-Platforms for the SKIF supercomputer program in 2004. The T-Platforms supercomputer system for TSU is expected to become operational in late 2006 or early 2007.

"The construction of the supercomputer system for TSU means that the CIS high performance computing market shifts to a new technological level," said Vsevolod Opanasenko, General Director of T-Platforms Company. "SKIF Cyberia is not just a computer, it's a finished world-class supercomputer center using the latest achievements of supercomputer technologies. We are grateful to our partners -- Intel and Microsoft -- for support and we are glad that leading Russian universities are ready to exploit the potential of high performance computing to its full extent."

Source: Based on information supplied by T-Platforms Company