Leslie D'Monte: On to next-level computing

BACKSTAGE

The world's fourth largest supercomputer "" and the most powerful one in the Asia and Asia-Pacific regions "" is now in India. The machine had to race against 500 other heavyweights from the world over to achieve this distinction. Called Eka, the Hewlett-Packard supercomputer was built in a record time of six weeks and at a cost of around $30 million (around Rs 118 crore).

The credit goes to a wholly-owned subsidiary of Tata Sons "" the Computational Research Laboratories (CRL). The main people behind the super achievement are N Seetha Rama Krishna (Project Manager), Sunil Sherlekar (Head, Embedded Innovation) and Ashwin Nanda (who heads CRL) besides, of course, Ratan Tata himself and CRL Chairman S Ramadorai, who is also the CEO & MD of Tata Consultancy Services (TCS).

The supercomputer has been installed at the CRL facility in Pune. CRL integrated this system with its own innovative routing technology and achieved 117.9 TFlop/s ('teraflops' or trillions of calculations per second) performance. It is three times slower than the world's fastest supercomputer "" IBM's Blue Gene/L which has a speed of around 475 TFlop/s and has been the numero uno since 2004. CRL, however, is already getting its petaflop-scale (1,000 trillion calculations per second) supercomputer ready.

"Getting into the top five itself has been a marathon exercise for us. Our next target is achieving the petaflop," asserts Krishna. It won't be easy though. Eka has 15,000 processors (CPUs), needs 400 tonnes of cooling, and occupies 4,000 square feet of space. "We faced multiple challenges. We had to connect the CPUs which we did with fibre optic cables (the first time in the world). This we did in the shortest duration (just six weeks), and the space which Eka occupies is one-and-a-half to two times less than what a similar machine would have required," he says.

Getting to the petaflop level would mean increasing the processor power by 10 times (150,000 CPUs), and the cooling to 4,000 tonnes, besides the increase in floor space. "All this requires a new architecture since growth in scale becomes non-linear at this stage," explains Krishna, adding: "Our plans are at the whiteboard stage." He should know. He has been in the field of high-performance computing (HPC) for the last 18 years, and has been developing multiprocessor systems to architecting end-to-end solutions in HPC and Grid computing domains. He has also been part of the C-DAC core team of HPC for 17 years which delivered PARAM 8000 to PARAM Padma.

Krishna points out that the key differentiating factor is that this system is not proprietary. It uses Linux, which is available in the public domain. "Any contribution done by the Tatas should go back to the public domain," he reasons.

Incidentally, it's for the first time ever that India has figured in the Top 10 Supercomputer Sites list. Supercomputers are used for highly calculation-intensive tasks such as problems involving quantum mechanical physics, weather forecasting, climate research, molecular modelling and physical simulations. CRL, for instance, used the system for aircraft modelling. The system ran for 17 hours on 168 CPU cores, thus saving on expensive wind-tunnel experiments.

"We have been in dialogue with the government and with other industry players but we did not get into any commitment as we wanted to have something in hand first," says Ramadorai. CRL is already working with Tata Elexis for animation work. Adds Sherlekar: "A lot of usage for this technology can come from within the Tata Group. We are in touch with Tata Technologies. Besides, TCS can also use it for commercial activities."