When Eka, the 117.9 teraflop supercomputer built by the Computational Research Laboratories (CRL) in Pune, India, was named the fourth most powerful machine in the world last November, the global computing community--and even the computer's developers--were surprised. Though a new top 10 list, due out this month, may see Eka demoted, CRL has already proved that top rankings in this field, typically the domain of national laboratories in the richest countries, cannot be taken for granted. Next month, the company should see its larger goals achieved as well: turning the US $30 million Eka into a revenue-generating supercomputer for hire and its team into a supercomputer services consultancy. Eka is the only privately funded supercomputer in the top 10, and it is the only one built specifically to make money.
”We believed that high-performance computing can earn money, be a profitable business, and make a difference to the economy of the nation,” says N. Seetha Rama Krishna, head of high-performance computing (HPC) operations at CRL, a wholly owned subsidiary of Tata Sons Limited, a division of the $28.8 billion-per-year conglomerate Tata Group, based in Mumbai.
True to India's software and services tech culture, rather than try to outdo Cray, IBM, Hewlett-Packard, or Silicon Graphics at designing and selling supercomputers, CRL will provide end-to-end supercomputing services--renting computer time, adapting and fine-tuning applications, and offering analytical services.
Today Eka is testing more than 15 applications for customers, and the company is in talks with several clients from the automobile, aerospace, financial, oil and gas exploration, and life sciences sectors, including aerospace giants Boeing, Embraer-Empresa Brasileira de Aeronautica, and Airbus.
”That commercial outfits are willing to bet huge money on HPC is surprising,” says S. Ramakrishnan, director general of the Center for Development of Advanced Computing, another supercomputing center in Pune. ”CRL could be a game changer in this arena.”
To get the job done, CRL started with a conventional cluster model of supercomputer, a group of 1800 Hewlett-Packard servers made up of more than 14 000 processors. The innovation lies in how the servers are connected to one another. The interconnect scheme relies on the concepts of projective geometry--a non-Euclidean method in which parallel lines intersect at infinity. The result is a simpler interconnect layout that both increases the computer's bandwidth and lowers its power consumption and cost of construction.
As a result of the interconnect scheme, the supercomputer is basically circular. At the center are data storage and networking hardware. Encircling those is the computing hardware, and surrounding it is a system of powerful air conditioners. This shape contrasts sharply with that of other cluster-type supercomputers, which are arranged as rows upon rows of computers.
The Eka team is still optimizing the machine, designing its own network switches and building specialized mathematical software. CRL has also begun work on the architecture and the key pieces of Eka's successor.