HPC R

1
HPC RD and applications in Chinas High-tech
Programme

• Depei Qian
• Beihang University
• June. 11, 2007, Edinburgh

2
863 effort on HPC

• Early 1990s from intelligent computers to
  parallel computers
• Intelligent Computer RD Center established
• 1993 Dawning 1, 640MIPS, SMP
• 1995 Dawning 1000, 2.5GFlops, MPP
• 1996 Dawning 1000A, cluster system
• 1998 Dawning 2000, 100GFlops, cluster
• 2000 Dawning 3000, 400GFlops, cluster,
  comercialized

3
863 effort on HPC

• 2002 Lenovo DeepComp 1800, 1TFlops, cluster
• 2003 Lenovo DeepComp 6800, 5.324TFlops, cluster,
  14th in TOP500
• 2004 Dawning 4000A, 11.2TFlops, 10th in TOP500
• 2004 Inspur high end enterprise servers T20000,
  up to 4TFlops

4
Other efforts on HPC

• Blade cluster (Galactic Computing)
• Clusters in universities
• Tsinghua University
• Shanghai University

5
Shift of emphasis

• From pursuing peak performance to pursuing
  scalability, usability, maintainability, and
  availability (SUMA)
• Commercialization
• Compatibility to main steam systems, supporting
  commercial software

6
Growing applications

• Applications from traditional areas (weather
  forecast, petro exploration) to much wider range
  of applications
• Bioinformatics
• Drug discovery
• Computational chemistry
• Materials
• Environment
• Manufacturing
• Simulation

7
Fast increase in HPC installations

• 2002 0.6 TFlops in TOP500
• End of 2006 94 TFlops in TOP500
• Terascale systems in major universities,
  institutes, and enterprises

8
Education/Training

• Training programme introduced in universities
• Practical training in application organizations
• Textbooks (translations) on parallel computing,
  algorithms and applications

9
HPC Environment

• 1999-2000 National High Performance Computing
  Environment (NHPCE)
• 2002-present CNGrid
• 10 sites
• 18 TFlops
• gt200TB storage
• 2003-present ChinaGrid (20 universities)

10
CNGrid nodes locations
11
Complementary RD in HPC Grid

• HPCs provide major resources to Grid
• Grid enables easy access to HPC resources and
  makes HPC available to a much wide user population

12
Complementary RD in HPC Grid

• Since 1998, 863 programme begin effort in HPC
  environment, resulting NHPCE, Grid term was first
  introduced
• 2002-2005 863 key project on HPC and core
  software, focused on Grid, resulting in CNGird
• 2006-present 863 key project on HPC and Grid
  service environment, significant upgrading of
  CNGrid resources, software, and applications

13
CAS Supercomputing Center (SCCAS)

• Equipped with Lenovo DeepComp 6800, 5.324TFlops
• Average utilization 86
• Applications
• gt180 registered users
• Supporting more than 120 important projects
• Important achievements

14
SCCAS User disciplines
15
SCCAS User distribution
16
SCCAS Projects supported
17
SCCAS simulation software OpenCFD
18
SCCAS bioinformation analysis toolkit
19
SCCAS Geo Science software GrADS
GrADS,providing interactive operation to the
geographical data. Applied in meteorological
applications.
20
SCCAS CNGrid Operation Center
21
Shanghai Supercomputing Center (SSC)

• Equipped with Dawning 4000A, 10.2 TFlops, disk
  95TB,
• Utilization 82.58

22
SSC Shared software
Software name source purpose
1 LS-DYNA com General purpose dynamics analysis
2 Fluent com CFD
3 MSC.Nastran com structural mechanics abalysis
4 MSC.Marc com advanced finite element
5 Feko com electromagnetics analysis
6 Pam-Crash com Crash and safety simulation
7 ANSYS Multi-physics com Finite element dynamics analysis
8 CFX com Flow dynamics
9 Grapes Self Global/regional weather forecast
10 NWChem OS Parallel computational chemistry
11 Siesta OS Molecule dynamics
12 Ego OS Molecule dynamics
13 CPMD OS
14 Abinit OS
15 Gromacs OS
23
SSC Users

• About 250 users
• Major portion Shanghai local
• But also from other parts of the country
• Users from university, institutes and
  enterprises, 1/3 each

24
SSC user distribution
25
SCC projects supported

• More than 200
• 863?973?NSF?Shanghai STC projects, etc.
• ARJ21 aircraft, outer ring tunnel aseismatic
  assessment, automobile industry

26
SSC Cross-Yangtze tunnel construction simulation

• Tunnel length 8.9 km,Diameter 15.2m

27
SSC Cross-Yangtze tunnel construction simulation
T20s
T40s
28
SSC Car safety analysis
LS-DYNA simulates car crash at 56km/h for 100ms
process, on Dawning 4000A, 32CPU, 3 hours
29
The new 863 key project

• Period 2006-2010, launched in Oct. 2006
• 640 million (RMB) from the MOST and at least
  equal amount of matching funding from different
  organizations
• Major efforts
• High productivity Computer development
• CNGrid development
• CNGrid GOS continuous development
• 12-15 Grid applications

30
Goal of the new 863 Key Project on HPC

• Developing petascale high productivity computer
• Innovation in HPC technologies, reaching the
  world level
• Advance in HPC design, manufacturing and
  applications

31
Major Challenges

• Novel architectures
• Memory/cache
• Parallel I/O
• Interconnect
• High performance computing node
• Low power consumption
• Reliability and robustness

32
Major Challenges

• Programming supporting
• Runtime support
• Scheduling of massive threads
• Multi-core related issues
• Algorithms for petascale systems
• System infrastructure
• Installation/configuration
• Cooling
• Fast deployment
• Application support

33
Two phase development of HPCs

• Two 100TFlops machines in the first phase (By the
  end of 2008)
• One 1000Tflops in the second phase (TBD)
• User?
• Investment
• Installation in CNGrid Major sites
• Promoting HPC and grid applications

34
Cooperation with UK/EU

• Cooperation with UK and EU HPC projects
• Researchers exchange
• Application experiences exchange
• Conferences/workshops
• Education/training
• Joint RD projects?

35
Thank you!