C-DAC 's Param programme sets to touch 10 teraflops by late 2007 and a petaflop by 2010.
THE Centre for Development of Advanced Computing (C-DAC), headquartered in Pune, has put India on the global supercomputing map. Alone among the multiple high-performance computing initiatives undertaken in India in the mid-1980s by scientific arms of the Central government (including the Council of Scientific and Industrial Research, and the Defence and Atomic Energy departments), C-DAC's Param project had a definite techno-commercial road map - and the centre managed to sell about half a dozen machines outside the country - and an equal number within it.
C-DAC developed the Param series of supercomputers as part of successive "missions" that culminated in the machine's entry into the global "Top 500" list of the world's most powerful computing platforms, in May 2003, with Param Padma, which was just shy of 1 teraflop (one trillion mathematical operations per second). In subsequent years, C-DAC went missing from the Top 500 list: the ever-shifting goal post of supercomputing ensured that a teraflop was no longer fast enough to make it to the supercomputing club.
The changing hardware scenario dictated that such number-crunching behemoths were replaced by large numbers of personal computers working in clusters to deliver the required performance. So instead of resting on its (teraflop) laurels, C-DAC swiftly readjusted its road map to the new hard reality of supercomputing: Grid computing was the new name of the game and the centre became the nodal agency for an ambitious countrywide grid linking all its own Param Padma platforms as well as some other computers in academia.
Garuda, the emerging network, is up and running - linking 15 powerful computers to the 1-teraflop mother machine housed in CDAC's National Param Supercomputing Facility in Bangalore, with a fast 100-megabits-per-second backbone.
The newest member of the supercomputing family is Param Sarita, a canny reworking of the basic computational punch to address the booming market for multimedia applications such as video-on-demand and interactive TV. All contemporary High Performance Computing (HPC) systems are scalable clusters of, typically, eight nodes, each fuelled by a state-of-the-art Intel Xeon chip.
The huge growth in bioinformatics - the harnessing of computers for drug discovery, protein modelling, deoxyribonucleic acid (DNA) studies - as well as the emergence of areas such as computational atmospheric sciences has created another opportunity for C-DAC's number crunchers.
The Computational Atmospheric Sciences (CAS) team at C-DAC is a key `customer' for Param's computational muscle. It engages in research on the application of high-performance computing to atmospheric and environmental science applications.
CAS has recently developed a `Real Time Weather System' (RTWS) capable of generating weather forecasts based on the user's choice. It is a Web-based, fully automated system that handles the complete weather forecasting cycle - from acquiring the data to its initial processing to simulating a model and intelligent post processing - all without user intervention.
It can be easily customised to handle data from a wide variety of sources and various numerical weather prediction models. The RTWS is useful for weather forecasters, atmospheric researchers and policy-makers. It can provide crucial support to fields such as aviation, oil, transport and sport.
Group Coordinator Akshara Kaginalkar explains that the system now makes use of C-DAC's Param infrastructure but has been designed to work on a variety of other computing platforms that the customer might have.
CAS is also involved in research on the Indian monsoon and uses global as well as regional coupled models for this in collaboration with premier national institutes.
For this, CAS has coupled meteorological models with air quality models and is carrying out simulations to study the impact of emissions and proposed control measures on the air quality in urban areas.
The HPC group is working round the clock these days to meet a self-imposed deadline: to assemble a supercomputing cluster, specifically tailored for biotechnology. Avionics design and gas exploration are some of the fields demanding new application arenas. It will be the first of Param's next-generation, high-performance, cluster-computing systems that will deliver more than 10 teraflops by the third quarter of 2007, explains S.P. Dixit, Director-in-charge, C-DAC, Pune. The hardware will be powered by state-of-the-art multicore processors; the interconnections are provided by C-DAC's PARAMNet-III.
The three-tier storage system will be scalable from 200 terabytes to a few petabytes. Like so many contemporary systems, the new Param will use reconfigurable elements based on field programmable gate arrays (FPGAs), which allow the hardware to be changed painlessly when required as the design evolves.
S.P. Dixit explains that this system is just a milestone on C-DAC's road map towards petaflop computing by the end of this decade. (A petaflop is a thousand teraflops, which is a million gigaflops, or a billion megaflops or mathematical operations.) And if that happens, it might just signal that C-DAC is back with a bang in the business of critical and high-end computing.
