As scientific enterprise increases in complexity, grid computing may be the answer
One of the antidotes to the dire prediction that the world of computing is reaching the limits of Moore’s Law, a thumb rule that predicts the expansion of computing power, is the deployment of computational grids. Pooling computer resources that are spread over geographic space and located in different administrative domains is now regarded as a cost effective alternative to deploying a single large machine at one location.
As scientific enterprise increases in complexity, as was dramatically demonstrated recently in the case of the Higgs Particle, collaboration is key. Grid computing is the deployment of an array of computers, which in their parent domains are used for a variety of purposes, but which can share their computational power to address “large” problems. The primary objective is to create what has been described as a “virtual supercomputer”. This is the reason why the field of computing has become synonymous with the network, said S. Ramakrishnan, the former Director General, Centre for the Development of Advanced Computing (C-DAC) at a meeting with “partner” institutions on Friday.
Socially useful purpose
Mr. Ramakrishnan, who was the founder director of ERNET, the national network of academic institutions in the country, observed that in order to socially justify the huge investments in such grids, it is essential to “demonstrate” their use to the public at large. He pointed out that the deployment of such grids in e-governance projects, which would impact a large number of people, is one way to achieve this. “We need to create a visible impact to justify the usefulness of such a grid,” he said. There are two essential aspects to the grid — a network of computers and a superhighway that enables the daisy chain to actually share resources. The Indian grid, GARUDA (Global Access to Resources Using Distributed Architecture) was deployed in 2004. The High Performance Computing (HPC) clusters were later chained to the National Knowledge Network (NKN), which brought many more institutions from across the country to the grid. The NKN’s highway now connects 774 institutional networks from across the country — universities, the labs of the CSIR and the DRDO, the Indian Council of Agricultural Research’s 42 institutions from across the country, nearly 100 medical institutions, the Indian Institutes of Technology, colleges and the ERNET itself.
C. Murali Krishna Kumar, Sr. Adviser (ICT), Planning Commission, said the objective is to extend the reach to 5,000 such institutions in a couple of years. “One-fourth of the field of computers should be touched by supercomputers in one way or other,” he remarked. He told The Hindu the government outlay for the grid could increase to Rs. 10,000 crore during the 12th Five-Year Plan.
Rajat Moona, Director-General, C-DAC, said the Union government has spent close to Rs. 5,000 crore on the grid. Observing that the grid currently has a computing capability of close to 70 teraflops (a trillion floating point operations per second), he hoped that Indian capacity would reach exascale capacities in three to four years.
The computing power of a supercomputer’s performance is measured in FLOPS (Floating Point Operations per Second). An exaflop is a quintillion, or a billion billion FLOPS. Computer scientists reckon that an exascale supercomputer could perform approximately as many operations per second as 50 million laptops. Currently, the fastest supercomputers operate in the petoflop range, a thousand teraflops. IBM Corporation’s supercomputer Sequoia, with a capacity of more than 16.3 petaflops, is said to be able to perform in eight hours the number of calculations the average laptop would take 20,000 years to complete.
What would you want to do with such a powerful machine? Mr. Moona says the computational power of such monster machines would be deployed on tasks such as climate and weather modelling, drug discovery, simulating molecular dynamics, the continuous monitoring of human bodies for predictive health analysis and many other tasks that require quick number crunching of large data sets. “The power of such machines would be in devices that would fit in our pockets in two decades,” he predicts. But what does it mean to have an exascale computer? It would do in minutes what the petaflop machines take days or even years, say computer experts.
R.S. Mani, Senior Technical Director, National Informatics Centre, the agency that executes the implementation of the grid, says the grid must be such that it is “scalable, adaptable and available full time”. The NKN is a hierarchical network divided into three basic layers — Core, Distribution, and Edge (User-Level). The NKN backbone has 18 Core Points of Presence (PoP) and about 25 Distribution PoPs across the country. Mr. Mani hopes that the superhighway would be extended to districts countrywide by 2013. A data centre in New Delhi and a disaster recovery centre (for data) in Hyderabad are to be established soon, he added.
“The big questions posed by big science and big data require collaboration,” which is what grid computing enables,” says Subrata Chattopadhyay, Chief Investigator, GARUDA. The Virtual User Communities on the GARUDA facilitate collaboration in a range of fields, from health informatics to bioinformatics and climate modelling.