C-DAC HPC Trends & Activities in · PDF fileC-DAC HPC Trends & Activities in India Abhishek...
Transcript of C-DAC HPC Trends & Activities in · PDF fileC-DAC HPC Trends & Activities in India Abhishek...
C-DAC HPC Trends & Activities in
India
Abhishek DasScientist & Team LeaderHPC Solutions GroupC-DACMinistry of Communications & IT Govt of India
Presentation Outline
A brief profile of C-DAC, India
HPC Activities & Trends
Challenges
Development Areas
C-DAC
is a premier national
Research & Development organization
under
Ministry of Communications and IT
Government of India
Set-up in 1988 with the explicit purpose of demonstrating India's HPC capability
Since then C-DAC has been undertaking building of multiple generations of
Supercomputer starting from PARAM with 1 GF in 1988.
As an institution for high-end Research and Development (R&D), C-DAC has been at the forefront of the Information Technology (IT) revolution, constantly building capacities in emerging/enabling technologies and innovating and leveraging its expertise, calibre, skill sets to develop and deploy IT products and solutions for different sectors of the economy, as per the mandate of its parent, the Department of Information Technology, Ministry of Communications and Information Technology, Government of India and other
stakeholders including funding agencies, collaborators, users and the market-place
MISSION
Provide leadership in comprehensive capability in HPC technologies and promote HPC as a problem solving tool in cutting edge research in Science and Engineering & grand challenge problems.
C-DAC Activities
End-to-End Solutions
Enabling Technologies
HPC, Language Computing, Speech Technology, e-Security, Geomatics,
Ubiquitous Computing Embedded Systems, VLSI, Broadband & Wireless,
Software Technologies, …
Science & Engineering, Strategic Sectors, Health, e-Governance, Education,
Power & Industrial Sector, Agriculture, Rural Areas, …
…
…
C-DAC HPC Evolution & Roadmap
Garuda – Grid
Computing
Social Computing
with participatory
approach
1991
1994
1998
2002-03PARAM Yuva
2008 54 TF
PARAM Padma
Viable HPC business
computing environment
PARAM 10000
Platform for User community
to interact/ collaborate
PARAM 8000
Technology Denial
2010
100 TF
2011
1 PF
PoC 100 Mbps
17 Locations
MainPhase
Garuda
PARAM 9000
June 1987
An Integrated Long Term Programme of Cooperation (ILTP) was signed between India
and Russia in Moscow for collaboration in Science and Technology areas.
Scope covers 12 thrust areas including Biotechnology and Immunology, Engineering Materials, Electronic Materials, Laser Science & Technology, Catalysis, Space Science and Technology, Physics and Technology of Accelerators, Water Prospecting, Computer and Electronics, Biomedical Science and Technology, Radio Electronics and Ocean Science and Technology, and
six selected areas of basic research in science, namely Mathematics, Applied Mechanics, Earth Sciences, Radio physics, Ecology & Environment, Chemical Sciences
and Biology.
PARAM Series….
With the areas identified for research collaborations, a parallel computing system, PARAM 8000 was first installed at Institute of Computer Aided Design(ICAD), Moscow in 1991-92 based on the then very powerful Transputer Processor Chip.
The Software dealing with Fluid Mechanics and Structural Analysis were operated and parallelized on the system.
National PARAM Supercomputing Facility
While the need and usefulness of High performance Supercomputing in Business as well as Scientific & Engineering Applications is unquestioned and is growing rapidly, it is not economically viable to have many such facilities.
C-DAC has thus established "National PARAM Supercomputing Facility“ (NPSF) to provide Supercomputing facility to various industries and other institutions that need such a facility to process their diverse applications.
PARAM Series housed at NPSF (PARAM 10000, PARAM Anant….)
GRID Garuda
The PARAM series of super-computers which are hosted at different geographical locations can be connected through GARUDA network
GARUDA is a collaboration of science researchers and experimenters on a nation wide grid of computational nodes, mass storage and scientific instruments that aims to provide the technological advances required to enable data and compute intensive science for the 21st century. One of GARUDA’s most important challenges is to strike the right balance between research and the daunting task of deploying that innovation into some of the most complex scientific and engineering endeavours being undertaken today.
The Department of Information Technology (DIT), Government of India has funded the Centre for Development of Advanced Computing (C-DAC) to deploy the nation-wide computational grid GARUDA. In Proof of Concept (PoC) phase which ended on March 2008, 17 cities across the country were connected with an aim to bring “Grid” networked
computing to research labs and industry.
PARAM Padma ranked 171
The PARAM Padma Supercomputing with peak performance of 1 TF in 2002 ranked 171th position in Top 500 list
India enters Supercomputing race
November 2007
IBM's BlueGene/L - 478 teraflops
IBM's BlueGene/P - 167.3 teraflops
SGI Altix ICE 8200 - 126.9 teraflops
HP Cluster Platform 3000 BL460c (EKA) - 117.9 teraflops
HP Cluster Platform 3000 BL460c - 102.8 teraflops
For the first time ever, India placed a system (EKA) in the Top 10. The Computational Research Laboratories, a wholly owned subsidiary of Tata Sons Ltd. in Pune, India
8 Indian supercomputers enter global top 500 list
By Siliconindia News Bureau
Monday, 08 December 2008
Tata Group's Eka, leading the race in the 13th rank
C-DAC)'s Param supercomputer at the 68th rank
Indian Institute of Science at the 213th rank
Industrial research company at the 334th rank
A research agency at the 428th rank
IIT-Madras at the 436th position
Paprikaas Interactive Services ranks at 478
These ranks were awarded by the Universities of Tennessee and Mannheim, Germany, and at NERSC Lawrence Berkeley National Laboratory.
India December 2009Summary
Total of 15 entries.
The performance criteria is minimum of 1.71 TFlops.
First 2 in the list are also in Top 500.
The average performance is 16.49 & combined performance is 247.28 TF.
11 of the systems are from HP, 2 from IBM, 1 from SUN, and 1 from SGI.
City-wise: Bangalore - 5, Pune - 3, Chennai - 3, Delhi - 2, Mumbai -1, Hyderabad - 1.
5 of the HPC systems are from (CDAC), proving its status as a leading high performance computing center in the nation.
PARAM Yuva
Compute Nodes Storage VisualizationCompilation
Nodes
Power & Cooling SAN 2 Hardware
Accelerator
NFS Data &
Management
Network
Hardware
Management
Network
SAN 1
Recent Trends
Many Cores onto a single die
Motivation is energy efficient performance
More & More Hybrid Clusters
Computational Nodes + Accelerators
Computational Nodes + GPUs
FPGA: Fixed functions, less flexible best performance/watt
GPGPU: Evolving towards general purpose computing, partially/ fully programmable
Recent Trends
RAM: Possibly magnetic RAM, faster Z RAM
Closer integration between processor & memory
Storage & Parallel File Systems
Interconnects improvements
Infiniband/ Optical/ Silicon photonics etc
New programming languages
Challenges
Application Performance Parallelism: How to Scale to 1,00,000 ways
Interconnects Effective Resource Utilization
Heterogeneity Application Portability and durability
Challenges
Application check pointing/ restart /migration Scalability
PowerGreen Components- Processors/Disks/Power suppliesAvoid several voltage conversions/ Power efficient UPS
CoolingPower efficient Chillers, using outside air cooling doors
Limit impact of failure to a subset of the system
Reliable components, Preventive maintenance and monitoring
Redundancy and Failover
Compute building blocks
Processors
RAM per core
GPGPU, FPGA .......
Operating System
Virtualization
Power & Cooling
Application Development
Porting and tuning
Algorithm development
Data partitioning and load balancing
Hybrid development tools & libraries
New Programming languages
Cluster provisioning & Management
Dynamic OS Provisioning
Support for Virtualization
Support for Heterogeneous and Hybrid Systems
Intelligent fault prediction
Load balancing and migration