Diffused goals

SUPERCOMPUTING MISSION: It is not a great idea to build a supercomputing grid fully based on imported systems and develop a community of users around

Diffused goals
The cabinet committee on economic affairs recently approved a National Supercomputing Mission with an outlay of Rs 4,500 crore spread over seven years. It is the first time in several years the government has committed such a level of funding for one single science-related mission.

In addition, it marks India’s return to investments in a strategic area of scientific activity – supercomputing – after a gap of more than two decades. The decision also emphasises continuity in government policies in science and technology. The supercomputing mission was originally conceived during the previous UPA’s tenure and formally announced by then prime minister Manmohan Singh in his address to the Indian Science Congress in January 2012.

India’s efforts in supercomputing have been rooted in the technology denial regime unleashed by America and other Western powers for decades. India was forced to develop its own supercomputer after it was denied one for weather forecasting by America on fears that it may be used for weapon-related research. Then PM Rajiv Gandhi, egged on by his Scientific Advisory Committee (SAC), approved establishment of a dedicated agency - Centre for Development of Advanced Computing based in Pune.

With its PARAM supercomputer unveiled in 1991, India joined a truly elite club which till then included only America and Japan. China, currently a global leader in supercomputing, joined the race a decade later but surpassed India soon. Sometime in 2010, SAC pushed the idea of reviving the supercomputer programme, resulting in the 2012 announcement by Manmohan Singh. The Planning Commission followed up by approving an outlay of Rs 2,000 crore for a National Supercomputing Roadmap for the 12th five year plan (2012-2017).

The revival of India’s supercomputer ambitions, however, got caught in turf wars and delays. This was despite the fact that CDAC already had a blueprint for the next level of development, and the country had a robust high performance computing community including a private sector player – Tata’s Computational Research Laboratories. But when the question of a new supercomputer project popped up, CDAC was overlooked despite it having all the necessary credentials. Instead, the Indian Institute of Science, Bangalore was chosen to steer the effort. The supercomputer centre in the city is involved in education and research relating to supercomputers but not so much in core development activity.

Subsequently, a compromise was worked out in 2013. The Department of Science and Technology and the Department of Electronics (DST) and Information Technology (Deity) were asked to jointly prepare an implementation strategy, in consultation with both IISc and CDAC. Now, the government says the mission will be implemented jointly by Deity and DST, through CDAC and IISc. The apex governing structure is likely to be headed by the PM.

Whatever may be the outcome of this exercise, India has already lost five years from the time the decision to re-enter the supercomputing race was taken in 2010. In 2007, both India and China had nine supercomputers each in the top-500 list. An Indian computer had made it to top 10.

Now, China is on the top and the ranking of Indian supercomputers has seen a slide. Losing five years in a critical and fast changing area of technology where the pecking order is revised twice every year is serious. Moreover, now we are talking of a seven year time-frame and not five years as proposed in 2010.

Unclear goals

The cause of worry is not just loss of time but also lack of properly defined goals for the proposed mission. In March 2013, the National Supercomputing Roadmap had identified “development of high capacity petascale supercomputers and technologies that may create an opening leading to building Exascale supercomputing capability” as its key objective. This would have meant a jump from teraflop stage to the next level of petascale and then aiming for the holy grail of exascale. Both America and China are in the race to develop an exascale supercomputer by 2020.

From the announcement made by the Narendra Modi government about the National Supercomputing Mission, the goal post appears to have shifted away from developing and building supercomputers to applications. A supercomputing grid of 70 high-performance computing facilities at national academic and R&D institutions has been proposed.

These supercomputers, which will be used for developing applications of national relevance, will also be linked to the National Knowledge Network. Instead of clearly spelling out if India will develop its own petascale machine or if it will invest in exascale development, the government has merely said that “the mission also includes advanced R&D” and that it will “create requisite expertise to build state-of-the-art next generation supercomputing.”

It is not such a great idea to build a supercomputing grid totally based on imported systems and develop a community of users around it. The Indian Meteorological Department is developing its own grid with imported hardware. The IISc has just acquired a petaflop supercomputer from Cray, making it India’s first such machine.

However, if the goal is to be a formidable player in the game, India needs to invest in developing both capacity and capability in building supercomputers as it did in the initial phase. Let’s not forget that supercomputing is still considered strategic and America continues to pull levers when it comes to supplies of crucial components to competitors. For the success of any scientific mission, both a defined goal and a deadline are must.

(The writer is Fellow, Centre for Media Studies, New Delhi)
Comments (+)