Resources

The creation of the supercomputer center for data processing is a very complicated task, because a big number of different hardware and software platforms (servers, disks, belt libraries, network computers, operational systems and the systems for capacity management, the facilities for reserve copying and safety facilities) must be united into one balanced system. Comprehensive in this case helps both to take into account the specificity of business-tasks of an enterprise and also optimize the compound and configuration of data and software platforms according to the specificity of the current and planned computational load.

Software and hardware complex of the Computing Center of URIIT can be used for fundamental and applied research, as well as for solving of a number of scientific and technical tasks.

First of all, these are the tasks concerning the developing of huge nature resources of KMAR. This region nowadays comes first in Russian Federation as to oil production, the second as to electric power output, and the third as to gas production. Exploring and developing of the interior of the require for the solution of complicated and resource capacious applied tasks of geoinformation, oil seismo-exploration, space information science, remote earth probing out of space, aeroplanes and satellites…

The resources of the Computing Center can be used for fundamental research in such important areas of present-day information science as the parallel computer architectures, distributed network calculation, neural networks, superscale databases and inverse problems of natural science.

The Computing Center of URIIT created specially for these problems is abreast with the most powerful world computing centers. At the same time it has flexible possibilities stage-by-stage development.

The core of the Computing Center

Computer core of the Computing Center consists of two powerful supercomputers by Sun Microsystems. The total maximum-demand power of these computers is more than 100 gigaflop, with the possibility of slight growth till hundreds of gigaflops at the expense of processor modules.
The most powerful computer of Sun Microsystems – Sun Fire 15000 is used as the platform for solving the most difficult computational tasks. This multiprocessor computer is made in SMP architecture. This architecture meets to the utmost the demands of many-dimensional computational tasks with tightly coupled data, to which many strategic tasks of URIIT relate – geoinformation, space information science, seismic and many others.
Supercomputer Sun Fire 15000 is built on the hierarchic principle, which allows great increasing of the computational power without the major loss of productiveness to interprocessor interaction. At the first hierarchical level there are the computational modules consisting of four processors UltraSPARC III with the clock rate of 900 MHz and on-line storage of 32 gigabyte, and also the in-out modules supporting Ethernet 10/100, Gigabit Ethernet, Fibre Channel, SCSI interfaces. Sun Fire 15000 is equipped with 18 slots for setting of computational modules and with 18 slots for setting of in-out modules, that results in maximum configaration of 72 processors and the memory of 576 gigabytes.

At the second hierarchical level computational and in-out modules are joined with the help of Sun Fireplane intercomponent commutation, that looks like three 18ơ18matrixez for independent commutation of data, addresses and controlling information. Peak commutation performance of 172,8 Gbyte/s and 43,2 Gbyte/s in ready-state regime provide non-disabling work of processors and memory modules, regardless of their position – either they are on one module or not. This balanced architecture allows the Sun Fire supercomputer to reach maximum productivity Rmax of 107,5 gigaflop, according to Linpack test, that is applied to this class of computers. This is 83% of supposed computer peak performance of 129,6 gigaflop. This is a very high efficiency in comparison to this class of supercomputers made by other producers, who usually use effectively no more than 60% of processor’s power.

Supercomputer Sun Fire 15000 is 99,999% reliable, at the expense of full hardware duplication, «hot» processor replacement, the possibility of non-stop service.

The dynamic domain architecture of this computer allows soft allocation of the resources between separate computational tasks and groups of tasks. Every domain, including several computational and in-out modules and representing virtual computer, is fully isolated from other domains. It provides the necessary level of reliability and the necessary flexibility is also retained. Dynamic domains controlled by network administrator allow to single out for each group of tasks of URIIT as much resources as they really need and to provide safety and independence.
Twelve computational modules and six in-out modules both meet the current demands of URIIT in computational resources and provide sufficient store for its further increasing and development.

The second supercomputer is cluster supercomputer built on the basis of 12 two-processors cross-points Sun Fire 280R and cluster software Sun Cluster 3.0. Two processors UltraSPARC III+ 900MHz Cu and two gigabytes of on-line storage are installed on each Sun Fire 280R server, as well as two hard disks with 36 GB Fibre Channel interface. Sun Fire 280R/Sun Cluster 3.0 allows solving effectively the tasks with weak data links that are supported through mutual disk drives. Cluster software Sun Cluster 3.0 provides both high reliability of supercomputer, due to the automatic activation of computational tasks of failed cross-point on the working cross-pint, and also the high performance regime due to simultaneous launching of several copies of tasks on several cross-points of cluster. It is very important that, if necessary, ome of domains of SMP-computer Sun Fire 15000 could serve as a claster crosspoint. This makes all the computational resources of the center as a common pool allocated between tsaks on common policy.

Forte software that is meant for the development of applications and also the number of dedicated libraries supporting parallel functions in these applications are used here.

The Software system Sun Grid Engine 5.2.2 allows URIIT specialists to develop specified applications, that can use any available computational resources, regardless of their location – either on domains of SMP computer, cluster cross-points or remote computers connected to computing center through local network or Internet.

Local network Gigabit Ethernet connects supercomputers with each other and also with peripheral devices such as graphical terminals and isolated servers. The technology of channel aggregation allows the exchange speed between the cross-point of 4 Gbit/s in duplex regime. The local network is fully dial-up. This provides its high productivity that is necessary for effective work of computation center in the whole.

Data storage network

Data storage network has three-level system of drives (disk massive, belt library and optic CD/DVD library) and allows to store safely tens terabytes of data. The drives are connected to supercomputers with the help of Fibre Channel commutations under redundant connections layout, that raises the productivity and makes the storage network refusal resistant. This progressive technology provides the high flexibility level of data storage devices. Any processor of SMP-computer and any cluster cross-point can have independent and high-speed to any disk. This allows applying different schemes of shared and separated access to disk, as well as to the data on belts and optic disks. At that the choice of scheme depends on software and the number of tasks of computing center.
For the time present SMP computer Sun Fire 15000 is connected to storage network. The cluster computer will be connected to this network later.

The general volume of data storage system is several tens of tetrabytes. And there is the possibility of data migration from one level to another in automatic mode.

Backup system

Backup system is based on Legato Company software. This company is the leading supplier of this class of software. The software is module developed. It allows implementing different backup schemes, archive and data migration, differing in their functional possibilities and productivity. The belt library of backup system has four belt drives LTO 100/200Gb and can hold about 160 cassettes. Thus the maximum volume of data storage is Tetrabytes. The server part of Legato software works on separated Sun Fire V880 server, assigned for database and important applications support. Sun Fire V880 server is connected to storage network FibreChannel and local Gigabit Ethernet network, that provides the access to all devices of computing center. The main backup regime is server less/LAN free. It means that the backup means have the access to devices without the participation of main computers, and also without loàding the main network. This regime completely unloads the computation cetnter components, processors and the means of network exchange, from any additional loading, that is usually caused by backup process, thus allowing to fully use them for their usual work.

Terminal system

Terminal system Sun RAY 150 is the centralised reliable solution for interactive interaction of computing center users with software processes in supercomputers. This terminal system consists of graphic terminals with keyboard and mouse, connected to local network with the help of embedded adapters. The central part of the system is the software working on one of servers of the center and supporting graphic interface between applications and hardware terminal. The centralised architecture allows to refuse from unreliable personal computers for graphic terminals and expensive X-terminals. To provide high safety the smart-card autentification of users is used.

Internet-portal

Internet-portal is made on a separate Sun Fire V880 server, connected to supercomputers by means of local Gigabit Ethernet network. Multiprocessor architecture, embedded disk subsystem and fault-tolerant components allow to use effectively this server for organisation of large Web-sites and Internet-portals, working with large databases and enduring intensive referencing. The server of portal is separated from Internet with the help of inter-network high-productivity screen. Oracle Application Server is used for interaction with the corporate database.

Management system

The management system Sun Management Center provides for the centralised direction of hardware and software of the Computing Center from one single control panel.