Supercomputer breakthrough from India

From the Hindu editorial on a supercomputer breakthrough achieved by Tata group:

For an indigenously built supercomputer to make it to the fourth spot in the world ranking of high performance computing (HPC) systems is a real breakthrough. For this, the band of researchers at the Tata Group’s Pune-based Computational Research Laboratory (CRL), who achieved this just 18 months after their facility was established, deserve the highest praise. While the Tata system has been built around off-the-shelf technologies, the top three spots have gone to entirely bought out proprietary supercomputing systems, such as the IBM’s BlueGene. A comparison of the private effort with the record of the Centre for the Development of Advanced Computing (CDAC), a public-funded supercomputing research facility, is unavoidable. Although CDAC made the first Indian teraflop supercomputer, PARAM Padma, in 2003, its successor is yet to achieve the kind of breakthrough in performance the Tata system has scored. The reasons for this include the ability of an innovative private enterprise to quickly mobilise resources to exploit the latest off-the-shelf processors and platforms to put together a world class HPC system.

Vijay Bharve, in his blog story has some more info and links. Hindu itself seems to have published the story a couple of days ago which (I, like, totally missed) describes some juicy politics (and in which IISc also makes an appearance):

Interestingly, however, the release from the Tata Group fails to make any mention of the main architect and prime mover of the project, Dr. Narendra Karmarkar, an alumnus of IIT Bombay and formerly of the Tata Institute of Fundamental Research (TIFR), who quit last year following lack of adequate financial support (to the tune of Rs. 400 crore) for his supercomputing project from the institute and joined the Tata Group having found the backing of Ratan Tata and the management of the group (see Business Line, May 6, 2006).

The CRL was established in July 2006 by a group of “like-minded alumni of IIT Bombay,” as the CRL website states.

However, earlier this year, Karmarkar left the Tatas (along with his core team) when they fell apart following differences over the overall plan and the set of goals and objectives of the HPC project, which included that the first system should be given to the Indian government.

But there can be little doubt that the basic idea of the architecture that has been used in the system that has been included in Top 10 belongs to Dr. Karmarkar.

According to Dr. Karmarkar, only five percent of his ideas, which he had shared with the Tatas, have been made use of the current architecture. “With even this, if it can make it to the fourth place, the system can surely make it to No.1 when all my ideas are incorporated,” he says.

“The design and concept took nearly four years, much of it was developed during my days at the TIFR. I have a long and complex plan of how to go about putting all my ideas into an HPC system. But I also have a balanced set of objectives. I am looking for a backer who would not only provide the money but should share the vision and the entire objective,” he added.

…

The other Indian systems include an IBM eServer Blue Gene Solution of the Indian Institute of Science (IISc), with 18.7 teraflops sustained performance and 23 teraflops peak performance and ranked 58th, another HP Cluster of Tata-CRL with 1440 processors with 9.3 teraflops sustained performance and 15.4 teraflops peak performance and ranked 179th , and six IBM systems in various industrial enterprises, ranked 152nd, 158th, 336th, 339th, 340th and 371st respectively.

I find it curious that the Hindu editorial which compares CDAC’s performance with that of Tata group fails to mention what Kamarkar himself has to say, namely, that

The design and concept took nearly four years, much of it was developed during my days at the TIFR.

Take a look!