C-DAC: Press Kit - Archives(In the News) - The Grid: Life beyond the Supercomputer

Last month India joined the league of nations boasting teraflop computing capabilities. Translated, a teraflop is an acronym for trillion floating-point operations per second, a measure of computing power.

Actually, supercomputing power. And the Indian supercomputer capable of powering along at a teraflop, we were told, was the PARAM Padma (capable of being scaled up to 16 teraflops), developed by Centre for Development of Advanced Computing (C-DAC).

Supercomputing applications mostly allow scientists to rip through computationally intensive tasks, like determining how the weather might be next week, or how a protein might fold in the course of designing powerful, new drugs. Supercomputing applications have already moved out from universities to corporate labs, but supercomputers aren't something you buy off the shelf. First, there's the danger of supercomputers being used for defense applications. With computing power like this on tap, you can simulate nuclear explosions too, a rather quiet, effective way of going thermonuclear.

That's why India's long been on a list of countries proscribed from buying US supercomputers. That's actually why the development of the PARAM series of machines continued through the 1990s. Then there's the cost. The PARAM Padma is priced at $5 million. If that sounds stunning, it's only half the cost of an equivalent supercomputer built in the West.

As you can see, supercomputers aren't for everyone. If all this sounds like an impressive achievement for India, here's a reality check. Government officials will tell you how there are 52 PARAM installations all over India, including four in Russia and one each in Germany, Canada and Singapore. The problem is that the earlier series of PARAM were built and used on the fly, much like early Tata cars. Six years ago, I remember being hugely impressed when I saw my first PARAM, towering over me at its home in Pune. But when I began running a check with its users (there were 25 installations then), I found the PARAM was an oddity. It was not used for serious computing work because its operating environment was too unfamiliar and cumbersome to users brought up on plug-and-play software. Like early Indicas, the PARAMs were experiments made on consumers.

I haven't seen the PARAM Padma, but when it becomes ready for use sometime this month, I'm not sure High-Performance Computing users will beat a path to its door. The reason for that is simple: sheer power is no longer the domain of standalone supercomputers, whose makers around the world have realized that teraflop speeds can now be achieved by lesser machines.

To my mind, Internet history is about to repeat itself in high-end computing. In the summer of 2002, the US National Science Foundation began installing the hardware for the TeraGrid, a transcontinental 'virtual' supercomputer created by networking clusters of lesser computers so tightly that they behave like one entity. The TeraGrid is capable of 13.6 teraflops, or more than eight times the most powerful academic supercomputer in the US. But more than speed, the TeraGrid signified the official acceptance of grid computing - the large-scale integration of computing systems to offer performance unattainable by any single machine.

The idea of grids has already spun out from academic in the US. Biotech giant Monsanto developed software - a key component in getting grids to work - that allowed its scientists to draw computing power from hundreds of Compaq and Sun Microsystems machines already in use at its offices. Grid software splits a problem into thousands of pieces that can be solved independently - say by an idle computer in an office - and stitched back together. Today Monsanto can, without supercomputers, do a gene analysis (of food staples like rice or wheat) in less than a day, a fifty-fold increase over what was possible five years ago. So it's not surprising that computer giant IBM embraced grid computing a year ago and now says all its products will come with grid software as a standard feature. Admittedly, setting up grids isn't as easy as it sounds. Getting many computers to work seamlessly isn't easy, and security is a big issue. But it can be done. Grids are particularly good for a country like India with limited resources but clusters of computers, many with long idling periods in academic institutions. Fortunately, while announcing the PARAM Padma, the government said it intended to set up an I-grid linking High-Performance Computing sites across India. That isn't a bad idea at all, but first let's just start linking up our lesser computers. It could give us a PARAM Padma at a fraction of the cost.