SPECIAL FEATURE: C-DAC -- HIGH-TECH ROAD TO DEVELOPMENT

Target: Teraflops

ANAND PARTHASARATHY

Among the supercomputing initiatives launched in India, C-DAC's Param sees techno-commercial fruition even as it puts India in the global `teraflop' club.

The C-DAC Knowledge Park in Bangalore, which houses the Terascale Supercomputing Facility.

The Centre for Development of Advanced Computing (C-DAC) was established in March 1988 as a scientific society of the Department of Information Technology (formerly Department of Electronics) of the Union Ministry of Communications and Information Technology (formerly Ministry of IT). Primarily a research and development institution involved in the design, development and deployment of advanced IT-based solutions, the two thrust areas for its first decade were supercomputing and Indian language computing. Over the years, C-DAC has diversified its activities to address the requirements in various areas - financial and capital market simulation and modelling, network and Internet software, health care, real-time systems, e-governance, data warehousing, artificial intelligence and natural language processing. In February 2003, the government announced the merger of the Electronics Research and Development Centre, India (ER&DCI), the National Centre for Software Technology (NCST), and the Centre for Electronics Design and Technology, India (CEDTI), Mohali, with C-DAC. The restructured C-DAC was expected to offer economies of scale and avoid duplication of work. In effect, the enlarged C-DAC has become a major R&D player, with an asset base of around Rs.220 crores and a staff strength that has more than doubled to 1,600. Its revenue is expected to be around Rs.100 crores: A feature of C-DAC, right from its inception, has been the entrepreneurial and techno-commercial thrust given to all its work, as a result of which it usually generated at least half of what the government spent on it. Some inevitable overlap in responsibilities have largely been adjusted and two years after the `merger' (which was effected from December 2002), "C-DAC Mk II" is just about ready to address more challenging tasks warranted by its new strength in human and infrastructural resources. In a new era driven by the Internet's global reach, C-DAC has the technological muscle to deliver on large national projects in the public interest that private industry may not always be able to address. In its 17th year, C-DAC is ready. The challenge is to harness its rich talent and resources in a sensible manner that will ultimately make a difference to the people of India. The following pages mirror its past achievements as well as its potential for future good.

WHEN C-DAC was born in 1988, four separate initiatives in supercomputing were being pursued in India. The pioneer was the Bangalore-based National Aeronautical Laboratory (now National Aerospace Laboratory), which even by 1986 had put together what was possibly the first parallel processing platform in India - the Flosolver.

In Delhi, Sam Pitroda had motivated the Centre for Development of Telematics (C-DOT) to put together its own supercomputing machine - CHIPPS (C-DOT's High-Performance Parallel Processing System).

The Department of Atomic Energy (DAE) and the Defence Research and Development Organisation (DRDO) had their own in-house compulsions to create large number-crunchers: The Bhabha Atomic Research Centre (BARC) began work that culminated in the Anupam supercomputer that Electronics Corporation of India Ltd (ECIL) produced in small numbers. The DRDO, driven by the need for advanced computational fluid dynamics (CFD) studies for its Light Combat Aircraft (LCA) project, set up a new unit in Hyderabad - Anurag - which created its own high-performance computer, PACE (Processor for Aeronautical Computations and Evaluation).

The Param Padma supercomputer in Bangalore.

All four initiatives were funded by the Indian taxpayer - so who wanted a fifth? The government of the day apparently wanted it - the memory of the Indian Institute of Science's (IISc) abortive attempt to purchase one of the leading American supercomputers, the Cray, purely for academic research, still rankled and it was a matter of national pride to be able to say, `If they won't sell it, we'll make it'. Which was why the development of yet another supercomputer was one of the major mandates of the fledgling C-DAC.

While the other supercomputing projects produced a few systems that were `given' to captive departments or institutions, it was the C-DAC venture - Param, Sanskrit for `supreme' - with no in-house customer to speak of, that worked towards techno-commercial success and found dozens of buyers in India and abroad.

In order to deliver supremacy in supercomputing, C-DAC operated in a mission mode. The First Mission for three years, with a Rs.32 crores in the kitty, aimed at creating a 1 gigaflop machine - that is a billion mathematical calculations a second. To achieve this, C-DAC created a 256-node parallel computing design and used the then most popular chip for such applications, the transputer. Massively parallel processing (MPP) was then the hot new architecture for the supercomputer, a move away from the classical techniques of creating a single number-crunching behemoth. Now they chopped up each computational challenge into a number of roughly equal tasks, then allotted them to dozens of separate but identical units that worked simultaneously, in parallel.

The first Param 8000 machine was beefed up with Intel's I 860 microchips to create the enhanced Param 8600. And when the SuperSparc chip became available, C-DAC found a replacement for the ageing transputers that kicked up performance of the Param 9000 to 5 Gflops (floating-point operations per second).

In 1993, Param's Second Mission was launched - a five-year game plan aiming for a 100 Gflop machine, the Param 10000. C-DAC learned the hard way that such speeds could not be achieved entirely with off-the-shelf components, so its engineers designed and built their own communication co-processor and also created their own communication and interconnecting hardware, named ParamNet.

The availability of meaningful application software is crucial to any new supercomputing platform that aspired to be marketable. C-DAC filled the gap by setting up a dedicated National Supercomputing Facility in Pune, with a Param 10000 and its variant Param Ananth and inviting researchers from dozens of scientific institutions to use the machines to try out their own applications in seismic data processing, structural mechanics, molecular modelling, weather forecasting and the emerging science of gene mapping. It was a canny move that paid rich dividends: of all Indian supercomputing machines, the Param developed a `fan club' of users who helped spread the skills to use it.

The Third Mission for C-DAC in 1999 had a relatively short time span - three-and-a-half years - and the task to lead the nation into the tera era. A teraflop machine cranked out data at a trillion (a million million) floating point operations a second. By the turn of the century, supercomputing architectures had evolved. Current thinking was that the best way to achieve high-performance was by adopting a `cluster' architecture. Not just hundreds of nodes in a single parallel processing machine, but hundreds of individual computers, all strung together to create a giant system. The economy came from the individual computers, which were fairly standard off-the-shelf desktop machines - and hence fairly cheap. C-DAC adapted itself to this change and for its teraflop platform, the Param Padma, it deployed 248 Power-4 chips in a 64-way cluster of IBM machines. It clocked just over 1 teraflop at peak speed.

On April 1, 2003, the Param Padma was dedicated to the nation in its brand new setting - the Terascale Supercomputing Facility within the C-DAC Knowledge Park in Bangalore. In June that year, it took India for the first time into the ranking of the world's Top 500 supercomputers (http://www.top500.org/lists/2003/06/). It was ranked 171st.

The latest Param was a 1,000 times more powerful than the first. In a span of 15 years, C-DAC had built and delivered four generations of supercomputing platforms, and over 50 of these machines were in active use in India as well as in Russia, Canada, Germany and Singapore among other countries. After the United States and Japan, India was the only country to have built supercomputers for use beyond its own shores.

As the next major step in the next four years, it would not only be faster computers but more of them linked to a grid that would drive the vision of C-DAC. The mission could not have been better timed when big countries and big players are focussing on the same, looking for inflection points. C-DAC has commenced another journey.

A national computing grid is not an easy job. But when the goal is achieved, it will be a truly awesome resource. And going by earlier record, this will be yet another mission where C-DAC will deliver.

Printer friendly page  
Send this article to Friends by E-Mail