India beats US curbs to produce supercomputer
BANGALORE: The one-teraflop (TF) scalable supercomputing cluster, claimed to be the largest machine in the Asia-Pacific outside of Japan, is capable of setting the pace for an Indian revolution in the field of bio-informatics, weather forecasting, fluid mechanics in space technology, seismic data processing in oil and gas exploration, drug discovery and computational chemistry.
"At present we cannot import systems because of export control restrictions. But we have sourced some of our components from elsewhere and developed our own technologies to produce Padma," R.K. Arora, executive director of the Centre for Development of Advanced Computing (C-DAC), told reporters here.
Currently, the fastest supercomputer, a 36 TF machine, exists in Japan, followed by the U.S. with 13 TF.
"But the 248 processors in Padma are scalable to 1,200, providing storage facility to go up from five terabytes to 22 terabytes. This is why we say it
is state-of-the-art," added Indian IT secretary Rajeeva Ratna Shah.
One TF, or 1,000 gigaflops (GF), is equivalent to one trillion floating point operations per second. For the layman, it means calculations that a normal computer would take six months to perform can be done in a matter of minutes on this supercomputing cluster.
About 200 scientists developed the Param Padma at a cost of 500 million over the last four years. The supercomputer has been displayed at the sixth international conference on high performance computing in the Asia Pacific region, which opened here Monday.
The interconnect switch -- the Paramnet II -- and the entire suite of systems software tools including management, debugging, compiling and engineering solutions have been developed indigenously, Arora said.
U.S. restrictions on India importing supercomputers came in the 1980s on suspicions that New Delhi was using dual use technology for nuclear purposes.
But the work of the scientific community at C-DAC as well as the Indian Institute of Science (IISc) here led to the production of the Param series of parallel computing -- the Param 8000 (one GF) in 1991, the Param 8600 (five GF) in 1993, the Param 9000 (10 GF) in 1995, Param 10,000 (100 GF) in 1998 and the Param Padma I 2002.
Seven pieces of the Param 10,000 have already been sold abroad, with Russia purchasing four and Canada, Germany and Singapore buying one each.
"We will look at the overseas market for Padma too, at least to those countries which need it. At the current rates, it should cost 50 percent lower at $5 million when compared with similar products from the developed countries," said Arora.