development of high performance computing on the basis of sscc and СС intel-sb ras development of...
TRANSCRIPT
Development of High Performance Development of High Performance Computing on the Basis of SSCC Computing on the Basis of SSCC
and and СССС Intel-SB RAS Intel-SB RAS
B.M. Glinsky, N.V. Kuchin (ICM&MG SB RAS)
V.V. Samofalov (Intel)
22
Competence CenterCompetence Center ( (CCCC)): Goals of : Goals of
CreationCreation
Introduction of modern computing technologies based on Intel developments and SB RAS achievementsinto science and industrial production of Siberian region
Training of organizations of mining industries, science, and higher education in modern computing technologies based on Intel developments;
Consultations on parallel programming and computing services on the basis of SSCC clusters (Intel Itanium2 platforms; prospects: multicore Quad-Core Intel® Xeon® 5300, etc.) on parallel supercomputations;
A comparative estimate of the efficiency of Intel new developments in the field of technical facilities and software on the basis of real problems solved in SSCC.
CC is being created in accordance with the Memorandum of Cooperation between SB RAS and INTEL of June 14, 2007
33
Important tasksImportant tasks
Realization of the goals set is of primary importance for scientific research and equipment of SSCC with modern SW of Intel.
Of no less importance is the problem of performing high-performance computations, consulting and computing services of SSCC and Novosibirsk Branch of Intel for large industrial enterprises of Novosibirsk and Siberian region.
Another important problem is the support of joint projects of Intel with some institutes of SB RAS (Institute of Inorganic Chemistry, Institute of Semiconductor Physics, Institute of Catalysis).
44
Tasks ofTasks of Intel – CC SB RASIntel – CC SB RAS
key concepts: infrastructure of the Data Processing Centerkey concepts: infrastructure of the Data Processing Center, , bladeblade--solutionssolutions, , multicore structuremulticore structure, , parallel calculationsparallel calculations
Creation of a community on high-performance computing (HPC Community);
Understanding of real problems of science and industry for which high-performance computations (development of algorithms and programs, industrial calculations, the use of commercial/free packages) are necessary;
Providing of computing resources for science, higher education, and industry on the basis of Intel HW&SW;
Consultations and practical assistance in solving problems of users; Participation in the organization of a Siberian segment of HPC Grid
(Novosibirsk, Tomsk, Krasnoyarsk); Increasing the practical efficiency of high-performance computing
facilities of Intel.
4
55
Methods to achieveMethods to achieveset goalsset goals
A Siberian SuperComputer Center of Resource Sharing (SSCC RS) has been created in SB RAS at ICM&MG• Remote access from 22 SB RAS institutes, 3 universities – Competence
Center base. • Possible access from other organizations via other communication
channels. • NCS-160 cluster on Itanium2 platform (ICM&MG, Intel Novo site, HP):
Pilot version of Intel (8 servers rx1620, InfiniBand); 82 servers and communication equipment have been purchased The center development is financed by the Presidium of SB RAS, Instrumentation Commission,
and RFBR grants. Joint Seminar of the Department of Computing Systems and Networks
of ICM&MG SB RAS, Computing Systems Department of NSU, and Intel Novo in the interests of the Competence Center• Series of seminars-trainings on HPC on SSCC RS resources performed by
specialists of Intel and SSCC for users from SB RAS, universities, and industry.
66
Creation of Competence Center:Creation of Competence Center:
Expected ResultsExpected Results Prompt mastering of a new generation of
computing technologies will:• Provide Siberian region with new computing
platforms for effective development of science and industry,
• Assist in the realization of joint SB RAS- Intel research developments.
Planned actions will make it possible to:• Train SSCC users in modern technologies of high-
performance computations, • Attract users to SSCC not only from institutes of
SB RAS but also from the industry of Novosibirsk and region.
77
Some resultsSome results
SSCC obtained single-user academic licenses for Intel SW A Work Group of CC Council consisting of representatives
from science and industry has been created. Chairman: RAS Corr. Member B.G. Mikhailenko
An agreement of cooperation of CC with SibNefteGeofizika and Chaplygin Siberian Research Institute of Aviation in the field of high-performance computing has been reached.
A contract with Schlumberger on the development of high-performance computing methods and computing services in the interests of oil industry has been made.
8
COMPUTING RESOURCESMVS-1000/32
Disc array МA8000
2,5 Tbyte,max – 3 Tbyte
Graphics workstationSunBlade 2000
VISUALIZATIONSYSTEM
ICM&MGnetwork
Internet NSCnetwork
SSCC network
EthernetGigabitEthernet (GE)
FibreChannel
GE
GE
GE
NCS-160 (Intel)
SSCC structureSSCC structure
SOFTWARE
MVS-1000/128
System
applied
mathematical SW
…………………………
164processors.Itanium 2,1,6 GHz,InfiniBand,GigabitEthernet,
мах – 1 TFlops
128 processors.Alpha 21264,667 MHz,
Myrinet,FastEthernet,
196 GFlops
32 processors.Alpha 21264,833 MHz,
Myrinet,FastEthernet,
50 GFlops
DATA STORAGE SYSTEM
9
Tasks of Tasks of gas-and-oil producing industrygas-and-oil producing industry((SchlumbergerSchlumberger))
Modeling of processes of acoustic wave propagation in the 3D case for problems of acoustic logging in wells
(ICM&MG SB RAS)
Modeling of acoustic logging in a well with a layered space
Velocity in the well: 1500 m/s;
Fast formation: P-wave: 3500 m/s, S- wave: 2500 m/s;
Slow formation (horizontal layer):
P- wave: 2500 m/s, S- wave: 1200 m/s.
On the left: symmetric location of the source
On the right: asymmetric location of the source (the source is at half radius).
10
Numerical modeling of seismic wave fields in 2D Numerical modeling of seismic wave fields in 2D inhomogeneous elastic media of different scalesinhomogeneous elastic media of different scales
((karstickarstic inclusions inclusions)) (Trofimuk Institute of Oil and Gas Geology and Geophysics)
Modeling, with the finite difference method, of objects of subseismic scale; calculation of wave fields emerging on karstickarstic inclusionsinclusions
The use of a fine grid for the entire region requires large computing power
Local mesh refinement in zones with subseismic inhomogeneities is proposed.
11
a)
a)
b)
b)
Vertical component of the full wave field. a) Uniform fine mesh
b) Mesh with local refinement in the zone of karstic inclusions The deviation of wave fields from each other is about 0.1%; demands for computational resources differ by about an order of magnitude
Horizontal component of the full wave field. a) Uniform fine meshb) Mesh with local refinement in the zone of karstic inclusions
12
ITAM SB RASITAM SB RAS
Simulation of flows around “Progress” space Simulation of flows around “Progress” space shipship
13
Modeling in a new developmentModeling in a new development of the spaceship “Clipper” of the spaceship “Clipper”
ITAM SB RASITAM SB RAS
Flow density distribution
Pressure distribution
Control effectiveness of the spaceship “Clipper” has been investigated.Its aerothermodynamic parameters at a descent from the orbit to 80 km have been determined.3D calculations were made by the method of 3D statistical Monte Carlo modeling taking into account chemical reactions.
About 100 hours of MVS 1000/128 processor time were used in the calculations.
Processor time distribution
of SSCC clusters (January-December 2007)
Science: SB RAS institutesEducation: Universities – NSU, NSTU, SSUTIIndustry: State Research Center of Virology and Biotechnology "Vector“, Novosibirsk Technology Center of Schlumberger
Time (CPU * hour)
NCS-160 MVS-1000/128
Science 377351,16 84,94% 363774,25 91,89%
Education 29233,97 6,58% 23919,87 6,04%
Industry 37692,36 8,48% 8177,63 2,07%
Total 444277,50 100% 395871,75 100%
Thank YouThank You!!
15