India bets a billion dollars to regain supercomputing clout
January 23 2012: In his address at the 99th Annual Session of the Indian Science Congress at Bhubaneshwar on January 3, Prime Minister Manmohan Singh remarked: “Over the past few decades, India's relative position in the world of science had been declining and we have been overtaken by countries like China”. To address this decline, he said, “We must ensure a major increase in investment in R&D, including by industry and strategic sectors….”
Among the measures that government was considering, Dr Singh mentioned a proposal the sheer scale and funding for which took many in the audience by surprise: “We are examining a proposal to build national capacity and capability in supercomputing which will be implemented by the Indian Institute of Science Bangalore at an estimated cost of Rs. 5000 crore.”
Rs 5000 crore – that is Rs 50 billion or $ 1 billion, is a very large sum of money -- possibly more than all previous indigenous supercomputing initiatives in the country put together, including the National Aeronautics Laboratory’s pioneering Flosolver, the Centre for Development of Advanced Computing’s Param series of supercomputers, the Defence R&D Organization’s PACE/ANURAG supercomputer the Bhabha Atomic Research Centre’s Anupam, the Institute for Mathematical Sciences’ Kabru Linux cluster, the Tata-funder Central Research Laboratory’s Eka supercomputer and most recently the Wipro build Saga 220 for the Vikram Sarabhai Space Centre of the Indian Space research Organization.
Versions of the Param; the IIMSc Kabru and the CRL Eka have figured at various times in the global Top 500 ranking of the fastest computing machines – though today the Eka remain the only indigenously assembled platform on the list, Param and Kabru having dropped out, with the shifting goalposts of giga flops and peta flops required to make the best 500 cut.
The scientific community has generally attributed India’s declining presence among supercomputing nations to the disinterest of successive governments and the tapering off of crucial funding to keep the main publicly funded high performance initiatives alive.
That may all be set to change – and with a kitty of a billion dollars, a new Indian supercomputer among the world’s top 100 or 200 machines looks doable.
The Prime Minister’s announcement has surprised many of the institutions and individuals who have been associated with the indigenous supercomputer development programme for a quarter century, since it appears to have overlooked the expertise available in these institutions in actually building a HPC machine as opposed to buying and running an imported system.
More recently a proposal for a national Exascale mission – a generation beyond today’s top-speed peta flop machines -- has already been submitted to government – and leaders of India’s most successful supercomputing programme to date – the Param series -- like Dr Vijay Bhatkar, are known to have helped craft the proposal. It mooted a short term budget of Rs 500 crores for 2 years though it also arrived at the same figure ( Rs 5000 crores for the total mission till 2020. But while this proposal seemed to suggest a sensible collaborative approach with the participation of all institutions with the required R&D expertise,the PM's speech seems to imply that the nation's supercomputer project is being sought to be entrusted to IISc alone. It might well be that a small reference might in a long speech might not explain the nuances of government's intentions. And IISc will in all probability not try to go it alone, but will sensibly leverage the expertise that lies outside the walls of its own Supercomputer Education and research Centre – but within India, at both government and private sector institutions.
Indeed that would seem to be not only sensible but the only way to go forward if India is to regain the position it once occupied in the small club of nations capable of building the world’s fastest computing machines.
S Ramakrishnan, veteran technology adminstrator and former Director General, CDAC, who steered the Param programme in many of its crucial years spells out what needs to be done: What is needed now, is to build all round national capability leading to the delivery of machines with petaflops of computing power in the coming three years and an exaflop machine ( 1000 petaflops) by 2020, with core competence in every aspect - hardware ( hybrid architecture with multi-core & GPU), system design, packaging, cooling, powering , (except perhaps the processors), memory and storage, interconnect, system software ( compilers, libraries, algorithms optimized for the architecture & hardware with multi-threading ), applications which leverage the underlying architecture. USEFUL LINKS AND BACKGROUND: PM's speech at the on 3rd January 2012 at Indian Science Congress ( see link among his speeches in left hand column) http://pib.nic.in/newsite/pmreleases.aspx?mincode=3
Ministry of Science and Technology Sanctions 5000 Crore for Supercomputer Research http://egov.eletsonline.com/2012/01/ministry-of-science-and-technology-sanctions-5000-crore-for-supercomputer-research/
Professor N Balakrishnan, Associate Director, IISc, and supercomputing project coordinator, said, “There is an urgency among scientists and policy-makers that India has fallen behind in supercomputing. India is currently ranked 140th in the world; it was ranked third in 2007. The fall is drastic and alarming and has to be arrested immediately. IISc and DST will work to revive the country’s original strength.”