Here come the "number crunchers"
EVERY DAY, in all weather conditions, scientists at the Indian Space Research Organisation (ISRO) receive data about the earth from spaceborne satellites that use specialised RADAR equipment. But converting these data into high resolution image of the earth is a mindboggling task: To process an image of a 100 km by 100 km area, it would take over 50 billion mathematical calculations.
And, in Bangalore, scientists at the Aeronautical Development Agency (ADA) in Bangalore can now simulate air-flow through aircraft engines they design, taking into consideration air pressure, friction and aircraft speed.
How do they do it? In both cases, the scientists are aided by supercomputers. And, both are stamped, "Made in India". The rocket scientists used PARAM, an indigenously built supercomputer, and the ADA team had access to the Bhabha Atomic Research Centre (BARC) parallel processing system (BPPS) in Bombay. The ADA team had the information it needed in three days. Previously, says an ADA scientist, "It would have taken at least 12 days of continuous computing on our imported IBM computer."
Supercomputers have made earlier methods of aerospace designing and flight simulation sound like grandmothers' tales. "Earlier, aircraft-engine testing was done in wind tunnels," says S C Purohit, a coordinator at the Centre for Development of Advanced Computing (C-DAC), a scientific society of the department of electronics (DoE) that developed PARAM in 1991. Operating a wind tunnel meant blowing wind at various velocities onto an aircraft model fitted with sensors and was a long, tedious and expensive process. The maintenance of wind tunnels cost millions of rupees annually.
Supercomputers are changing all this and more. They are entering the fields of medicine, satellite image processing, industry, academics and even crime fighting.
>>> C-DAC and the Sanjay Gandhi Post-Graduate Institute of Medical Sciences in Lucknow are working on a system to enable doctors to calculate radiation doses by looking at 3-D images of patients' bodies. Doctors will also be able to store, restructure and analyse X-rays, magnetic resonance imaging and scan images.
>>> Indian Space Research Organisation units in Hyderabad, Bangalore and Thiruvananthapuram use supercomputers for processing satellite images to search for natural resources like petroleum.
>>> DoE scientists in New Delhi are trying to find out how supercomputers can be used in massive industrial controls.
>>> The Maharashtra police has approached C-DAC to find out the feasibility of using supercomputers to maintain their voluminous data on crimes, criminals and fingerprints.
>>> University students are beginning to use supercomputers to study complex molecules and analyse biomedical images collected from video cameras and microscopes. The Indian Institute of Technology (IIT) in Delhi uses a supercomputer to demonstrate how a million circuits can be packed onto a square-cm chip.
The third mode
In the years ahead, supercomputers will play an increasingly important role in India's (R&D). Says C-DAC executive director V P Bhatkar, "Supercomputing is now widely recognised as the third mode of scientific investigation, complementing theory and experimentation."
Supercomputer technology is only a couple of decades old. And, just as telescopes can see the light of stars that are yet to reach the earth, supercomputers can provide a clear view of the future.
Environmentalists, for example, often make prophecies based on supercomputer images. Greenhouse gases and pollutants affect the environment in myriad ways, seriously affecting water, crops, forests and ultimately, our daily lives. Such complex interconnections are understood and analysed better by producing computer models of world economies and ecosystems. The US National Centre for Supercomputing Application, along with other agencies, is involved in such programmes.
In the US, the National Oceanic and Atmospheric Administration and the National Centre for Atmospheric Research are trying to predict global warming trends in the next century. The atmospheric research centre recently predicted how agriculture in Asia and North America would suffer if fossil fuels continued to be burnt at the present rate.
Supercomputers came into use in the 1960s to aid the construction of climate models and weather prediction -- an area in which India is now making steady progress. Later, the advent of supercomputers like the Cray-YMP helped scientists make more accurate predictions in more fields.
India's history of supercomputers begins with the Cray supercomputer and weather prediction. The country's first supercomputer was the Cray-XMP, imported in the mid-1980s from USA for the National Centre for Medium Range Weather Forecasting (NCMRWF) in New Delhi, and operated under US supervision. However, in 1989, the US rejected a request for a second Cray for the Indian Institute of Science (IISc) in Bangalore. Fearing India would use supercomputers for space and atomic research, the US banned their export to India. This decision prompted Indian policy makers and scientists to make indigenous supercomputers.
In their search for an alternative, Indian scientists resorted to parallel processing, instead of the conventional, high-speed, vector supercomputing. Parallel processing uses several processors to share the workload simultaneously (See box). An oft-quoted allegory on parallel supercomputing makes it clear: "It is a lot easier to harness 100 horses than to breed one that is 100 times stronger."
India's supercomputing success serves as an example of its growing self-reliance. Never mind that it was prompted by the West's denial of technology, which was displayed again recently when the US scuttled Russia's transfer of cryogenic rocket engine technology to India, forcing the country's scientists to work on indigenous engines.
But as India begins to finds its place in the sun, the US appears to be softening its stand. Recent news reports indicate the US trade representative's office has initiated moves to review the situation. Says N Vittal, former DoE secretary, "The US may become more lenient and make computers of higher performance available." But he cautioned, "In the light of our experience with the cryogenic engine deal, it is better for us to remain self-reliant."
Scientists are talking in terms of great leaps in supercomputers. The race now is to achieve better speed, sharper memory, faster communications and wider applications. And, an increasing number of people -- from university students to neurosurgeons -- are eager to see what these "number-crunchers" can offer.
C-DAC has upgraded its PARAM-8600 to a peak speed of 5 billion floating point operations per second and has embarked on a five-year project to achieve a peak speed of one trillion floating point operations per second (See box). "The world's fastest supercomputers -- the high-powered, liquid-cooled, sequential ones like Cray -- are still in the range of 20 billion or so calculations per second," scientists point out.
Says Bhatkar, "We are entering the field of frontier research." There are only three other competitors in the field -- the US, a European consortium and Japan, none of which has achieved a speed of trillion operations per second. "Our computers are already on par with European ones," claims Bhatkar.
Others, especially BARC, are equally confident, claiming better performance in specific fields and at less cost. The cost of the BPPS-64, a 16-node (consisting of 16 parallel processors) system would be about Rs 3 crore by next year, but its performance would be comparable to Western supercomputers that cost 10 times more. "We started the supercomputer project when we were convinced we could make a machine better than PARAM," says H K Kaura, head of BARC's computer division. "And we did it."
Supporting BARC's claim are ADA and IISc, which jointly tested BPPS-16. ADA scientists confide the BPPS-16 clearly outperformed PACE, PARAM and ADA's imported IBM. ADA scientist Kota Harinarayana says, "We have big plans for BPPS," adding that ADA also uses C-DAC's machines.
Meanwhile, the Hyderabad-based Advanced Numerical Research and Analysis Group (ANURAG), a division of the Defence Research and Development Organisation (DRDO), claims that its supercomputer PACE (Processor for Aerodynamical Computation and Evaluation) with 128 nodes and priced at Rs 35 lakh, is one of the cheapest in the field. The National Aeronauticals Ltd (NAL), which began work on supercomputers in 1986, says its Flosolver, though only half as powerful as the Cray-XMP, costs less than one-tenth the amount (Rs 2 crore).
The government has accorded top priority to supercomputer research. Says Vittal, "The government has allocated Rs 48 crore for C-DAC's second mission." The department of atomic energy is giving high priority to developing BARC's supercomputer system.
There are also moves to popularise supercomputers. Says Vittal, "We had initiated a Rs 78-crore World Bank-aided project to upgrade the computing capabilities of 32 institutions and make supercomputers available at college level.
Several mainstream computer and electronics firms now have plans to market them. Computer scientists point out the ultimate market success for supercomputers would be to make them useful for buyers. "PARAM has a whole range of functions for a large number of users," says R Srinivas, an R&D engineer of C-DAC who doubles as a sales executive. He and other engineers go around the country to meet potential buyers.
The advantages of parallel supercomputers are that they are cheaper and flexible. The 256-node PARAM costs Rs 2 crore, but the 64-node machine comes for Rs 55 lakh and the popular 16-node variety costs Rs 20 lakh.
The lower cost has made parallel processors a favourite in the world market. "Not only are they cheaper," points out Kaura, "it is also the technology of the future. Besides, parallel computers require no special cooling systems."
Kaura explains the difference between parallel and vector processing is not always watertight. He says Intel, the computer giant, is developing chips -- the building blocks of computers -- in parallel. Cray uses parallel processors in its vector computers. Indian supercomputers, such as BPPS and PARAM-8600, also use Intel's "i860" processors.
However, the drawback of parallel computing is the slow development of its specialised software, mainly because the technology is new, say scientists. The market for indigenous supercomputers will undoubtedly be influenced by software availability and options. "As of now, end-users often develop or modify software to suit their needs," says a C-DAC engineer.
"Buyers expect us to supply software off-the-shelf," says C-DAC engineer Sriram. "We are ready to develop software suited for the buyers' needs, but that kind of R&D culture is lacking in India."
On the other hand, vector supercomputers have an edge because they use readymade software. "Cray has an advanced software available. It is a big advantage," points out S S Prasad, deputy director of NCMRWF.
Another drawback of parallel processing is the multiplicity of supercomputer languages, which makes computer networking difficult. For example, C-DAC's PARAM uses its own system software and modifications of FORTRAN and C as languages. But, in tune with developments in Indian supercomputers, collaborative programmes are under way between users. Prasad says networking will be an important feature in the years ahead. At IIT in New Delhi, efforts are on to adopt and modify programmes used in the Cray to suit parallel processors.
Meanwhile, the latest DST annual report notes: "Planning for the replacement of the existing Cray-XMP/216 supercomputer has already started in view of the long lead time required." The report also notes the consensus of a recent review meeting on supercomputer trends. The report concludes: "The consensus that emerged from the session was that parallel processing systems have not yet become proven tools for medium range weather forecasting within the country or abroad. Therefore, research and development have to be fostered in the country in this area." -- With inputs from Koshy Cherail