C-DAC HPC Trends & Activities in India - HPC Advisory Council
C-DAC HPC Trends & Activities in India - HPC Advisory Council C-DAC HPC Trends & Activities in India - HPC Advisory Council
C-DAC HPC Trends & Activities in India Abhishek Das Scientist & Team Leader HPC Solutions Group C-DAC Ministry of Communications & IT Govt of India
- Page 2 and 3: Presentation Outline � A brief pr
- Page 4 and 5: � Set-up in 1988 with the explici
- Page 6 and 7: C-DAC Activities Science & Engineer
- Page 8 and 9: June 1987 � An Integrated Long Te
- Page 10 and 11: National PARAM Supercomputing Facil
- Page 12: PARAM Padma ranked 171 � The PARA
- Page 16 and 17: 8 Indian supercomputers enter globa
- Page 18 and 19: C-DAC latest Supercomputing Facilit
- Page 20 and 21: PRSG - March 2010
- Page 22 and 23: PARAM Yuva
- Page 24 and 25: Application drivers on PARAM Yuva
- Page 26 and 27: First 4 TF HPC Facility at North Ea
- Page 28 and 29: Recent Trends � RAM: Possibly mag
- Page 30 and 31: Challenges � Application check po
- Page 32 and 33: Interconnect � Bandwidth � Topo
- Page 34 and 35: Cluster provisioning & Management
C-<strong>DAC</strong> <strong>HPC</strong> <strong>Trends</strong> & <strong>Activities</strong> <strong>in</strong><br />
<strong>India</strong><br />
Abhishek Das<br />
Scientist & Team Leader<br />
<strong>HPC</strong> Solutions Group<br />
C-<strong>DAC</strong><br />
M<strong>in</strong>istry of Communications & IT<br />
Govt of <strong>India</strong>
Presentation Outl<strong>in</strong>e<br />
� A brief profile of C-<strong>DAC</strong>, <strong>India</strong><br />
� <strong>HPC</strong> <strong>Activities</strong> & <strong>Trends</strong><br />
� Challenges<br />
� Development Areas
C-<strong>DAC</strong><br />
is a premier national<br />
Research & Development organization<br />
under<br />
M<strong>in</strong>istry of Communications and IT<br />
Government of <strong>India</strong>
� Set-up <strong>in</strong> 1988 with the explicit purpose of demonstrat<strong>in</strong>g <strong>India</strong>'s <strong>HPC</strong> capability<br />
� S<strong>in</strong>ce then C-<strong>DAC</strong> has been undertak<strong>in</strong>g build<strong>in</strong>g of multiple generations of<br />
Supercomputer start<strong>in</strong>g from PARAM with 1 GF <strong>in</strong> 1988.<br />
� As an <strong>in</strong>stitution for high-end Research and Development (R&D), C-<strong>DAC</strong> has been at<br />
the forefront of the Information Technology (IT) revolution, constantly build<strong>in</strong>g capacities <strong>in</strong><br />
emerg<strong>in</strong>g/enabl<strong>in</strong>g technologies and <strong>in</strong>novat<strong>in</strong>g and leverag<strong>in</strong>g its expertise, calibre, skill<br />
sets to develop and deploy IT products and solutions for different sectors of the economy,<br />
as per the mandate of its parent, the Department of Information Technology, M<strong>in</strong>istry of<br />
Communications and Information Technology, Government of <strong>India</strong> and other<br />
stakeholders <strong>in</strong>clud<strong>in</strong>g fund<strong>in</strong>g agencies, collaborators, users and the market-place
MISSION<br />
Provide leadership <strong>in</strong> comprehensive capability <strong>in</strong> <strong>HPC</strong><br />
technologies and promote <strong>HPC</strong> as a problem solv<strong>in</strong>g tool <strong>in</strong> cutt<strong>in</strong>g<br />
edge research <strong>in</strong> Science and Eng<strong>in</strong>eer<strong>in</strong>g & grand challenge<br />
problems.
C-<strong>DAC</strong> <strong>Activities</strong><br />
Science & Eng<strong>in</strong>eer<strong>in</strong>g, Strategic Sectors, Health, e-Governance, Education,<br />
Power & Industrial Sector, Agriculture, Rural Areas, …<br />
End-to-End Solutions<br />
Enabl<strong>in</strong>g Technologies<br />
<strong>HPC</strong>, Language Comput<strong>in</strong>g, Speech Technology, e-Security, Geomatics,<br />
Ubiquitous Comput<strong>in</strong>g Embedded Systems, VLSI, Broadband & Wireless,<br />
…<br />
…<br />
Software Technologies, …
C-<strong>DAC</strong> <strong>HPC</strong> Evolution & Roadmap<br />
1991<br />
1994<br />
1998<br />
PARAM 9000<br />
PARAM 8000<br />
Technology Denial<br />
PoC<br />
100 Mbps<br />
17 Locations<br />
2002-03<br />
Ma<strong>in</strong><br />
Phase<br />
Garuda<br />
Garuda – Grid<br />
Comput<strong>in</strong>g<br />
Social Comput<strong>in</strong>g<br />
with participatory<br />
approach<br />
PARAM Padma<br />
Viable <strong>HPC</strong> bus<strong>in</strong>ess<br />
comput<strong>in</strong>g environment<br />
PARAM 10000<br />
Platform for User community<br />
to <strong>in</strong>teract/ collaborate<br />
PARAM Yuva<br />
2008 54 TF<br />
2011<br />
1 PF<br />
2010<br />
100 TF
June 1987<br />
� An Integrated Long Term Programme of Cooperation (ILTP) was signed between <strong>India</strong><br />
and Russia <strong>in</strong> Moscow for collaboration <strong>in</strong> Science and Technology areas.<br />
� Scope covers 12 thrust areas <strong>in</strong>clud<strong>in</strong>g Biotechnology and Immunology, Eng<strong>in</strong>eer<strong>in</strong>g<br />
Materials, Electronic Materials, Laser Science & Technology, Catalysis, Space Science<br />
and Technology, Physics and Technology of Accelerators, Water Prospect<strong>in</strong>g, Computer<br />
and Electronics, Biomedical Science and Technology, Radio Electronics and Ocean<br />
Science and Technology, and<br />
�six selected areas of basic research <strong>in</strong> science, namely Mathematics, Applied<br />
Mechanics, Earth Sciences, Radio physics, Ecology & Environment, Chemical Sciences<br />
and Biology.
PARAM Series….<br />
� With the areas identified for research collaborations, a parallel comput<strong>in</strong>g<br />
system, PARAM 8000 was first <strong>in</strong>stalled at Institute of Computer Aided Design<br />
(ICAD), Moscow <strong>in</strong> 1991-92 based on the then very powerful Transputer<br />
Processor Chip.<br />
�The Software deal<strong>in</strong>g with Fluid Mechanics and Structural Analysis were<br />
operated and parallelized on the system.
National PARAM Supercomput<strong>in</strong>g Facility<br />
� While the need and usefulness of High performance Supercomput<strong>in</strong>g <strong>in</strong> Bus<strong>in</strong>ess as<br />
well as Scientific & Eng<strong>in</strong>eer<strong>in</strong>g Applications is unquestioned and is grow<strong>in</strong>g rapidly, it is<br />
not economically viable to have many such facilities.<br />
� C-<strong>DAC</strong> has thus established "National PARAM Supercomput<strong>in</strong>g Facility“ (NPSF) to<br />
provide Supercomput<strong>in</strong>g facility to various <strong>in</strong>dustries and other <strong>in</strong>stitutions that need such<br />
a facility to process their diverse applications.<br />
� PARAM Series housed at NPSF (PARAM 10000, PARAM Anant….)
GRID Garuda<br />
� The PARAM series of super-computers which are hosted at different geographical<br />
locations can be connected through GARUDA network<br />
� GARUDA is a collaboration of science researchers and experimenters on a nation wide<br />
grid of computational nodes, mass storage and scientific <strong>in</strong>struments that aims to provide<br />
the technological advances required to enable data and compute <strong>in</strong>tensive science for the<br />
21st century. One of GARUDA’s most important challenges is to strike the right balance<br />
between research and the daunt<strong>in</strong>g task of deploy<strong>in</strong>g that <strong>in</strong>novation <strong>in</strong>to some of the<br />
most complex scientific and eng<strong>in</strong>eer<strong>in</strong>g endeavours be<strong>in</strong>g undertaken today.<br />
� The Department of Information Technology (DIT), Government of <strong>India</strong> has funded the<br />
Centre for Development of Advanced Comput<strong>in</strong>g (C-<strong>DAC</strong>) to deploy the nation-wide<br />
computational grid GARUDA. In Proof of Concept (PoC) phase which ended on March<br />
2008, 17 cities across the country were connected with an aim to br<strong>in</strong>g “Grid” networked<br />
comput<strong>in</strong>g to research labs and <strong>in</strong>dustry.
PARAM Padma ranked 171<br />
� The PARAM Padma Supercomput<strong>in</strong>g with peak performance of 1 TF <strong>in</strong> 2002 ranked<br />
171th position <strong>in</strong> Top 500 list
<strong>India</strong> enters Supercomput<strong>in</strong>g race<br />
� IBM's BlueGene/L - 478 teraflops<br />
�IBM's BlueGene/P - 167.3 teraflops<br />
�SGI Altix ICE 8200 - 126.9 teraflops<br />
November 2007<br />
�HP Cluster Platform 3000 BL460c (EKA) - 117.9 teraflops<br />
�HP Cluster Platform 3000 BL460c - 102.8 teraflops<br />
� For the first time ever, <strong>India</strong> placed a system (EKA) <strong>in</strong> the Top 10. The<br />
Computational Research Laboratories, a wholly owned subsidiary of Tata Sons<br />
Ltd. <strong>in</strong> Pune, <strong>India</strong>
8 <strong>India</strong>n supercomputers enter global top 500 list<br />
By Silicon<strong>in</strong>dia News Bureau<br />
Monday, 08 December 2008<br />
� Tata Group's Eka, lead<strong>in</strong>g the race <strong>in</strong> the 13th rank<br />
� C-<strong>DAC</strong>)'s Param supercomputer at the 68th rank<br />
� <strong>India</strong>n Institute of Science at the 213th rank<br />
� Industrial research company at the 334th rank<br />
� A research agency at the 428th rank<br />
� IIT-Madras at the 436th position<br />
� Paprikaas Interactive Services ranks at 478<br />
�These ranks were awarded by the Universities of Tennessee and Mannheim, Germany,<br />
and at NERSC Lawrence Berkeley National Laboratory.
� Total of 15 entries.<br />
<strong>India</strong> December 2009<br />
Summary<br />
� The performance criteria is m<strong>in</strong>imum of 1.71 TFlops.<br />
� First 2 <strong>in</strong> the list are also <strong>in</strong> Top 500.<br />
� The average performance is 16.49 & comb<strong>in</strong>ed performance is 247.28 TF.<br />
� 11 of the systems are from HP, 2 from IBM, 1 from SUN, and 1 from SGI.<br />
� City-wise: Bangalore - 5, Pune - 3, Chennai - 3, Delhi - 2, Mumbai -1,<br />
Hyderabad - 1.<br />
� 5 of the <strong>HPC</strong> systems are from (C<strong>DAC</strong>), prov<strong>in</strong>g its status as a lead<strong>in</strong>g high<br />
performance comput<strong>in</strong>g center <strong>in</strong> the nation.
C-<strong>DAC</strong> latest Supercomput<strong>in</strong>g Facility
PRSG – March 2010
PRSG – March 2010
PARAM Yuva<br />
Compilation<br />
Nodes<br />
Compute Nodes Storage Visualization<br />
Power & Cool<strong>in</strong>g SAN 1<br />
SAN 2 Hardware<br />
Accelerator<br />
NFS Data &<br />
Management<br />
Network<br />
Hardware<br />
Management<br />
Network
PARAM Yuva
PARAM Yuva
Application drivers on PARAM Yuva
Multi-discipl<strong>in</strong>ary Applications
First 4 TF <strong>HPC</strong> Facility at North Eastern Zone
Recent <strong>Trends</strong><br />
� Many Cores onto a s<strong>in</strong>gle die<br />
Motivation is energy efficient performance<br />
� More & More Hybrid Clusters<br />
Computational Nodes + Accelerators<br />
Computational Nodes + GPUs<br />
� FPGA: Fixed functions, less flexible best<br />
performance/watt<br />
� GPGPU: Evolv<strong>in</strong>g towards general purpose<br />
comput<strong>in</strong>g, partially/ fully programmable
Recent <strong>Trends</strong><br />
� RAM: Possibly magnetic RAM, faster Z RAM<br />
Closer <strong>in</strong>tegration between processor &<br />
memory<br />
� Storage & Parallel File Systems<br />
� Interconnects improvements<br />
Inf<strong>in</strong>iband/ Optical/ Silicon photonics etc<br />
� New programm<strong>in</strong>g languages
Challenges<br />
� Application Performance<br />
� Parallelism: How to Scale to 1,00,000 ways<br />
� Interconnects<br />
� Effective Resource Utilization<br />
� Heterogeneity<br />
� Application Portability and durability
Challenges<br />
� Application check po<strong>in</strong>t<strong>in</strong>g/ restart /migration<br />
� Scalability<br />
� Power<br />
� Cool<strong>in</strong>g<br />
Green Components- Processors/Disks/Power supplies<br />
Avoid several voltage conversions/ Power efficient UPS<br />
Power efficient Chillers, us<strong>in</strong>g outside air cool<strong>in</strong>g doors<br />
� Limit impact of failure to a subset of the system<br />
� Reliable components, Preventive ma<strong>in</strong>tenance and<br />
monitor<strong>in</strong>g<br />
� Redundancy and Failover
Compute build<strong>in</strong>g blocks<br />
� Processors<br />
� RAM per core<br />
� GPGPU, FPGA .......<br />
� Operat<strong>in</strong>g System<br />
� Virtualization<br />
� Power & Cool<strong>in</strong>g
Interconnect<br />
� Bandwidth<br />
� Topology<br />
� Latency<br />
� Protocols<br />
Fat Tree, 2D/3D Torus<br />
Hyper cube<br />
Hybrid
Application Development<br />
� Port<strong>in</strong>g and tun<strong>in</strong>g<br />
� Algorithm development<br />
� Data partition<strong>in</strong>g and load balanc<strong>in</strong>g<br />
� Hybrid development tools & libraries<br />
� New Programm<strong>in</strong>g languages
Cluster provision<strong>in</strong>g & Management<br />
� Dynamic OS Provision<strong>in</strong>g<br />
� Support for Virtualization<br />
� Support for Heterogeneous and Hybrid Systems<br />
� Intelligent fault prediction<br />
Load balanc<strong>in</strong>g and migration
Thank You