Friday, September 25, 2009

Indian IT: From Artisans to Global Industry

INDIAN CHEMICAL ENGINEER Volume 51, SPECIAL ISSUE, pp. 157-161

Indian Software: From Artisans to Global Industry

Shivanand Kanavi
Vice President (Special Projects), Tata Consultancy Services, 4th and 5th Floor, P.T.I. Building, No. 4, Sansad Marg, New Delhi - 110 001, India. Email: shivanand.kanavi@tcs.com

It is a great pleasure to contribute to this Festschrift in honour of Dr. Keki H. Gharda. He is a pioneer who innovated in process industries. India’s current global position in agrochemicals and generic drugs owes a lot to technocrats like him who used innovative process routes to produce high quality specialty chemicals. I am fascinated by him since I first met him in 1994 while researching for a cover story in Business India titled “India’s Technology Leaders” [1].

Today, India has achieved an undisputable leading position in the global software services. Most outsiders think that this has been primarily achieved due to labour arbitrage or availabilityof low cost computer programmers in India as compared to the US and Europe. However, that is only one of the contributing factors. The primary factor has not been noticed yet by many business historians. In fact, software development used to be an essentially artisan like activity about four decades ago. This paper argues that innovative industrialisation of such an individualistic activity to large-scale industrial activity by Indian firms has led to the current pole position of India in the field. To illustrate this thesis, the author will use the example and experience of Tata Consultancy Services (TCS) because he is familiar with it and also because being the oldest Indian software company it has also been an innovative pioneer.

Snake Charmers and Software

In 1968, when TCS came into being, there was no Microsoft, Apple, SAP, Oracle, Sun Microsystems, Dell and many others. Hewlett Packard (HP) was known for its oscilloscopes, signal generators, handheld calculators and not computers. EDS and Cap Gemini had just come into being as data processing companies.

Clearly, it was an audacious startup from India.

TCS had a couple of IBM machines of 1401 vintage and an ICL machine as well. They were providing bureau services, better known today as business process outsourcing (BPO), services for Indian telephone companies, power utilities, universities and so on. At the same time TCS also started scouting around for any work it could get abroad.

In those days India had gone through a severe financial crisis, the Rupee had been devalued by 57.5%. The US Dollars required for importing large and expensive mainframe computers were definitely in short supply. Besides, there was also a perception amongst the public and the politicians that computers would take away jobs and, hence, were best avoided. However, TCS leadership was convinced that computers could play a major role in not only making Indian businesses and banks more efficient but also in helping much of Indian society leap frog from the 19th to the 21st century. Surely, many people thought that the dream was incredible, but they kept chipping at the walls relentlessly. How TCS has been able to contribute to nation building in India through these efforts is a story to be told another time.

Collaborate and Flourish

In the mid-1970s, TCS decided that it would be good to get into an alliance with a computer maker. The one they chose was Burroughs Corp. Though Burroughs was not the biggest name in computer industry their technology was impressive. The deal was that TCS would sell Burroughs computers in India and support them. This helped TCS engineers gain first hand expertise in operating their systems and troubleshooting. TCS also bought a system for their computer centre to provide Burroughs based services to their clients. Burroughs soon realised that they had hit a gold mine of engineering talent and started outsourcing software work to TCS. Burroughs’ clients in Europe, US etc., who were switching their systems from some other manufacturer to Burroughs, would need their already running software to run on the new Burroughs machines. In those days each machine had its own dialect of COBOL and, hence, unless one carefully mapped the old programs onto the new Burroughs-understandable instructions, they would not work. TCS executed some interesting and challenging migration assignments in this mold.

TCS set up a strong Systems Group, which took the lead in assimilating technology as well as developing new ones. This Group came up with a tool called Data Dictionary, which helped automate software migration considerably. Burroughs was impressed. It tried its best to convince the Tata Group for a buyout, but the latter decided against selling out TCS and instead agreed to form a new joint venture, Tata Burroughs Limited. Interestingly, IBM, which was a competitor of Burroughs, was also impressed with TCS tools. TCS continued to execute migration and development work for Burroughs and at the same time look for other clients. Later, TCS decided to invest in IBM’s technology and started the most sophisticated IBM based computing centre in Asia of that time at Chennai. It was risky and expensive, moreover TCS had to convince the Department of Commerce, US Administration in Washington, D.C. that it was not going to use IBM computers to develop nuclear bombs, but in fact help US businesses become more efficient!

Henry Ford and Toyota

An important achievement of TCS has been the successful industrialisation and globalisation of software services. It has many similarities and a few dissimilarities with what Henry Ford and Toyota did to manufacturing. This has enabled TCS to execute large projects successfully. Year after year TCS has been delivering software solutions in targeted time and budget with a greatamount of certainty.

TCS did not have a model to follow, it had to invent one. Forty years ago computer programming was practiced by a few exceptional individuals. Each programmer had her/his own way of doing things, many of them brilliant but hardly replicable. It was difficult to debug or improve a program written by someone else. Naturally, it was almost never a product of teamwork. Obviously, it could not be scaled up. The situation was very similar to that faced by the auto industry when Henry Ford and his peers were developing quadricycles in their workshops in the 1890s. These workshops could not scale up and serve a mass market. That is why Ford’s assembly line, for Model-T, was pioneering and revolutionary in 1913. It was a product of meticulous planning and hard work in the background and it changed manufacturing forever [2].

Today TCS has achieved the depth and breadth to follow the same processes and achieve the same high quality and deliver them from any of our centres, be they in Hangzhou, Budapest, Sao Paolo, New Jersey, Toronto, Tokyo, Melbourne or in different parts of India. In Clayton Christensen’s terminology it would no doubt qualify as a “disruptive business model” [3].

Innovation Engine

Like any grand narrative it would be presumptuous to say that four decades ago TCS saw clearly its goal of industrialising software services, then charted the strategy, the path and, eventually, planned and executed to reach where it stands today. Like all big things, TCS started small with powerful ideas like Data Dictionary, a migration tool. TCS was involved in the software engineering standards and quality movement with IEEE from the very beginning. It also had to set up appropriate training of recruits, when there were hardly any colleges teaching computer science in India. TCS had to develop and set up processes to test and debug software.

In the late 1980s, TCS executed a large challenging project to set up the clearing and settlement system for banks in Switzerland. It was won against competition from established companies, purely on the basis of TCS’s innovative design. The project helped TCS “push the envelope” in all directions and also helped hone its software design and architecture skills as well as develop a core group of software architects. TCS also developed the systems required to integrate its client site work with the work done by teams of developers in India, known as ‘off-shoring’ and so on.

In 1981, TCS set up its R&D centre, Tata Research Design and Development Centre (TRDDC)in Pune [4]. A strong group in software engineering took shape at TRDDC in the mid-1980s. The group started articulating and evangelising concepts of software engineering, some of which were already in practice within TCS. This group was able to develop a highly successful suite of Computer Assisted Software Engineering (CASE) tools and carried forward the work initiated by TCS Systems Group in a methodical way. At another level, TCS also internally evangelised theSystem Engineering approach to software problems.

The agenda now involved identifying reusable components, knowledge repositories, creating a software tool foundry, developing highly sophisticated reverse engineering tools for software maintenance. The jigsaw pieces of industrialisation of software development started falling intoplace. One of the byproducts of this industrialising process was the development of the highly acclaimed MasterCraft™ – a suite for automatic code generation once the business logic is fed into it [4].

Setting Standards

On the front of standards TCS intensified its earlier work with IEEE and brought in SEI’s CMM philosophy into the organisation quickly. TCS is the world’s first organisation to achieve an enterprise-wide Maturity Level 5 on CMMI® and P-CMM® based on the most rigorous assessment methodology.

TCS has now combined its own vast store of home-grown processes with the best aspects of global standards, such as the SCMM, the PCMM, Six Sigma, ISO 9001 and the Tata Business Excellence Model, to develop its own proprietary quality model, the Integrated Quality Management System (iQMS™). TCS hopes that this archetype will soon become an industry standard.

The iQMS™ is central to project management at TCS; it comprises a major chunk of its DNA. This system provides guidelines for the conduct of every project and the means for monitoring it. Together with the various software development methodologies laid out by TCS’s software engineering process groups, iQMS™ lays out a comprehensive roadmap for each project. TCS has ensured that all its development centres, be they in China, Hungary, Chile, Brazil, Uruguay,US, Canada, UK, Singapore, India, Australia, Japan etc., follow the same processes and achievethe same high quality.

Learning from Manufacturing

TCS could not have set up the software equivalent of Henry Ford’s assembly line, if it did not build an efficient supply chain. In the 1960s and 1970s TCS started working in close collaboration with the newly set up Indian Institute of Technology (IIT) at Kanpur, Bombay and Madras and later expanded the company’s academic interaction to over 200 engineering colleges in India and several universities abroad. Today, TCS’s Academic Interaction Programme covers a whole spectrum of activities from faculty development; curriculum development in some colleges; scholarship and financial aid to deserving graduate programmes as well as sponsored research and collaborative development of Intellectual Property.

In India, this programme has contributed to raising the standards of computer science and software engineering education. As a result, TCS could recruit over 100,000 high quality engineers in the last four years and move them through its strong internal training programme that covers everything from software engineering to soft skills. The spade work and internal systems have helped to absorb this large human resource into the organisation quickly and deploy them into large projects.

To make TCS operations more efficient it setup a system meant for a global services company,very similar in concept to ERP in manufacturing. This system covers ‘everything’ – project billing, employee services, leave, pay roll and other HR services, internal communication, branding, online appraisal processes, knowledge management etc. Youngsters in TCS (incidentally, the average age in TCS is 26 yrs) thought that such an ultimate tool should be named Ultimatix, a la RenĂ© Goscinny and Albert Uderzo.

Truly Global Networked Delivery

In global manufacturing, it is well recognised that Toyota has taken the assembly line revolution of Henry Ford to the next level by introducing distributed manufacturing, Just in Time manufacturing, Single Minute Exchange of Dies (SMED) etc., which collectively have come to be known as the Toyota Model [5].

Similarly, TCS had to take its software factory approach to the next level. As its client list and diversity grew across continents, the company could not remain an India-based software developer serving global clients. About a decade ago TCS realised that it needed to further globalise its software development system, which became possible due to the global telecom revolution. The solutions and products offered by TCS are in bits and bytes and not in steel or aluminum. Hence, in some ways, TCS could venture into uncharted territory where a brick and mortar manufacturer like Toyota could not physically go.

Initially, TCS created the hub-and-spoke system in global delivery where India was the huband other centres were feeders. Today, however, TCS has gone further ahead to an entirely news ystem where any of its major global development centres could act as a hub or an anchor for a global collaborative effort. This leads to reduction of time zone issues for clients; facilitates services in a variety of languages and cultures; allows real time collaboration and parallel development with teams sitting in distributed development centres and so on. It also brings out optimum utilisation of in-house expert resources. TCS clients also enjoy de-risking greatly from putting all the eggs into the basket of one centre. Incidentally, TCS already has over 10,000 non-Indianemployees.

The TCS Global Network Delivery Model™ is the business equivalent of what Paul Baran proposed in his work entitled Introduction to Distributed Communications Networks for Rand Corporation, in 1964. It later became the conceptual framework for packet switched networks with no single centre, and no single path, like Arpanet and, ultimately, the Internet [6].

The pioneering conceptual and systemic work done by TCS percolated to other Indian IT companies in the last 10-15 yrs and lo and behold, India now has a vibrant, highly competitive, and high quality software services industry! The author has been asked many times about an order of magnitude difference in size between software companies in India and China. Even IBM, Accenture and HP have very large development centres in India with tens of thousands of Indian engineers. Are the engineering human resources in China any less in quantity than in India? The answer, as we all know, is no. However, Chinese software companies lack scale and the largest among them have less than 5,000 engineers. The reason is simple: they have yet to master the industrialisation of software development.

Many observers have pointed out the role played by English education, mathematical and analytical abilities among Indian students, propensity of Indian youth towards science and engineering as careers, labour arbitrage etc., as the determining factors in the rise of Indian IT industry. No doubt all these factors have played a role.

However, all such deterministic analysis ‘forgets’ the human factor of leadership. It is similar to saying that if a country has steel, gasoline and machinists then they will have a vibrant auto industry! Without the genius and hard work of Henry Ford and Toyota, the global auto industry could not have reached the scale and sophistication that it has today. Similarly, without the development and adoption of software engineering methodology in TCS and other Indian companies, global software services would not be a vibrant large scale industry that it is today.

References
1. Business India, July 4-17 (1994), http://reflections-shivanand.blogspot.com/2009/09/dr-keki-ghardaprofile.html
2. Ford, H. and Crowther, S., My Life and Work, Garden City Publishing Company, Inc., New York, USA (1922), www.gutenberg.net/etext/7213
3. Christensen, C.M., The Innovator’s Dilemma: When New Technologies Cause Great Firms to Fail, Harvard Business School Press, (1997).
4. Kanavi, S. (Ed.), Research by Design: Innovation and TCS, Rupa & Co. (2007).
5. Liker, J.K., The Toyota Way: 14 Management Principles from the World’s Greatest Manufacturer, McGraw-Hill Professional (2004).
6. www.rand.org/about/history/baran.html

Tuesday, September 22, 2009

State Bank of India: Story of its tranformation






Quaternion, September 15, 2005

A Colossus transforms

State Bank of India is a great story of transformation in the making. The implementation of TCS’s centralized banking solution in nearly 14,000 branches of the group will play a major role in it, reports Shivanand Kanavi.


If you ask Arun Kumar Purwar, Chairman, State Bank Group, about Operation Vijay, ("Operation Victory"), you will not get much in return other than a question shot back, "Where did you hear about it? That is still under wraps." But when strategy is being discussed with branch managers, who number in five figures, it would be difficult to keep it a secret. Apparently, it is a plan to transform the banking colossus in a tangible, quantifiable way. "It would be tall talk now. When we are close to achieving the targets, we will talk about it," says he.

Simply put, while the bank's position in Indian banking is unassailable, it wants to be number one in other financial services as well, be it credit cards, mutual funds or investment banking. "We have the fourth largest life insurance company of our country, the third largest merchant bank, the seventh largest mutual fund, the fourth largest credit card company, and the third largest factoring company. We have strategic interests in the credit information business and in an asset reconstruction company. We also run, within the group, 40-plus regional rural banks. Thus, State Bank's presence in the economy - in the financial sector - is very strong," adds Purwar.

Operation Vijay wants to take it to the next level. When stock market analysts call State Bank of India (National Stock Exchange: SBIN) as a 'play' on the Indian economy, in their own frenetic lingo, they are reflecting what SBI stands for. The bank, with its massive network of nearly 14,000 branches, is rivaled by none for its reach,
except a couple of Chinese banks. However the numbers cited by the Chinese are a little suspect because they list points of sales, including extension counters, as branches.

SBI controls an enviable 25% of the entire banking business in the country and services a whopping 100 million accounts. The bank has a wide network of 5,400 automatic teller machines (ATMs) in nearly 1,800 centers, and another 1,000 are being added this year. It includes an ATM at the highest altitude, at over 12,000 ft above sea level, in Leh, Ladakh, south of Tibet. At the other extreme SBI has an ATM on a river boat in Kerala. "Mind you, it is not a cash dispensing machine, but a fully networked ATM," assures Purwar. In January 2004, the total number of transactions from the ATMs was 6 million, and the total money drawn, between $138 million and $149.5 million (Rs 6 billion and Rs 6.5 billion). ATM usage is growing at 12% to 15% per month, according to Purwar. In fact, by Jan 2005, the total number of transactions on ATMs increased to 270 million, and money drawn amounted to over $ 644 million (Rs28 billion).

The largest competing network has less than 2,000 ATMs. Thus market analysts are justified in calling it a proxy for the Indian economy. That is, if the bank is doing well, then the Indian economy is doing well and vice-versa. (See box: The Bank)

The Bank
It is said that three organizations touch vast majority of the people of India: India Post, the Indian Railways and the State Bank of India.
The State Bank of India (established in 1806) is one of the oldest banks in the world and predates many well known names in the banking world: Citibank (established in 1812), Chemical Bank (established in 1823), ANZ (established in 1830), Standard Chartered (established in 1853), HSBC (established in 1865}, and the Bank of America (established in 1874).
Today, the State Bank of India, along with its nine associate banks form the State Bank Group, which does over 25% of all banking business in India through a network of 13,767 branches in India and 54 overseas, and over 5,400 ATMs in India's metros, and urban and rural areas. The group serves over 100 million accounts and, in the last four years, deposits have grown with a compound annual growth rate (CAGR) of 12% from $72.01 billion (Rs3,514 billion) in 2001 to $115.72 billion (Rs5,061 billion) in 2005. It was the only Indian bank to be listed among the top 100 banks of the world by The Banker in July 2005.
The shares of the banks are listed in the Indian stock exchanges and its global depository receipts are listed in Europe. SBI has a market capitalization of about $10 billion (Rs.441 Billion) on the Indian bourses (as of September 7, 2008). The Government of Indian owns 59.73% of the shares, through the country’s central bank, the Reserve Bank of India. It is the only bank in Asia, other than the Bank of Japan, to have a rating, from international agencies, above the sovereign.
The State Bank is a financial powerhouse with its arms in Insurance, mutual funds, investment banking, etc, and owns:
• SBI Capital Markets Ltd.
• SBI Mutual Fund (a trust)
• SBI Factors and Commercial Services Ltd.
• SBI DFHI Ltd.
• SBI Cards and Payment Services Pvt. Ltd.
• SBI Life Insurance Co Ltd – Banc assurance (Life Insurance)
• SBI Funds Management Pvt. Ltd.


The bank came under pressure in the late 1990s, with the opening up of the economy and competition from private banks. These new banks have no legacy issues, are lean and mean and could build themselves on hitech platforms and thus provide highly competitive services to both corporate and individual customers. But they lack reach and have largely confined themselves to the metros and large cities.

A Bicentennial
This year, SBI began celebrating it 200 years of existence. It traces its origin to the Bank of Calcutta, started in 1806 with the active participation of the British East India Company as a joint stock company. From it founding until 1936, the Bank acted as a central bank and currency issuing authority. In 1809, the bank received it charter and was re-designed as the Bank of Bengal. The Bank of Bombay (founded on April 18, 1840) and the Bank of Madras (founded on July 1, 1843) followed the Bank of Bengal. These three banks remained at the apex of modern banking in India. They came into existence as a result of the compulsions of imperial finance and the felt needs of local European commerce. They were amalgamated as the Imperial Bank of India on January 27, 1921.
When India attained freedom in 1947, the Imperial Bank had deposit of $ 63.273 million (Rs2.751 Billion), a network of 172 branches, and more than 200 sub-offices all over the country.

The commercial banks of the country, including the Imperial Bank of India had, till then, confined their operations to the urban sector and were not equipped to respond to the emergent needs of the rural areas, in order to serve the economy, in general, and the rural sector, in particular, the State Bank of India was created by taking over the Imperial Bank of India and integrating with it the former state-owned or state-associated banks. An Act was passed in Parliament in May 1955, and the State Bank of India was constituted on July1, 1955.
The Bank’s museum in Kolkata is a treasure house for economic historians, with detailed balance sheets and minutes of board meeting dating back two centuries.

"It was clear to us that either we transformed, or perished," says Purwar. But transforming is easier said than done. After all, it had 300,000 employees, systems that were not standardized, and some practices that are peculiarly Indian. The advantages of introducing advanced IT solutions with centralized databases, good networks, 'anytime, anywhere' banking through ATMs and so on, were clear to the leadership. "New technology would lower our transaction costs by 15%-25%," says Purwar. The bank's service level would rise with core banking. Its ability to monitor funds would dramatically improve with management information systems and daily reports. The available funds could thereby be deployed better in the money market, and the NPAs (short for non-performing assets, or bad accounts) could also be tracked in time, and action taken. New financial product-could be developed with what-if analyses, test marketing and deployment in record time. Technology would also enable the bank to enter new businesses and fee based services as the economy entered a low interest rate regime, with decreasing spreads for banks.

When old habits resist change, it takes leadership to drive change. The story is far from over but the signs of change are visible and indicate the potential this institution has. Today, over 4,800 branches are networked in a dedicated network called SBI Connect. Besides data communication, the network has vastly reduced the bank's telecom bills, as all these branches are using Voice Over IP phones for inter-branch communication. Over 5,400 fully networked ATMs in nearly 1,800 centers are bringing modern banking to even small towns and rural areas. Over 5100 branches of the SBI Group have gone live on core banking (a centralized database system). "In Belapur, we have our core banking systems, where over 800 officers are working hard to meet deadlines for the rollout," says Purwar. "A village may not have regular electricity and proper telecommunications, but our branch there will be computerized and connected to core banking," adds he.

And all this has been achieved with no downsizing! The number of branches on core banking in October 2004 was hardly 250. If the number today stands at 4,300, it works out to roughly 400 branches going live per month!

That is a truly amazing number by any standards. No wonder The Banker magazine, in London, honored SBI with its Technology Awards 2005, for the core banking project, and as the Best Outsourcing Project of the Year.

The project was outsourced to TCS as the prime system integrator. TCS developed a customized solution for SBI's needs. The implemented solution includes B@NCS from Financial Network Solutions (FNS), and Eximbills from China Systems interfacing with each other and the existing legacy system.

The building in Belapur, which houses the core banking servers in the data center and training facilities for the bank's staff, also has 300 TCS engineers working day and night.

That is where Krishnan Ramanujam, of TCS, who is the Director of the Core Banking project, sits. The blackboard outside his office tells it all. The numbers scrawled on it seem to change every time you visit him. If you did not know better, you would think he was running an illegal numbers racket out of his office. On closer inspection, however, they look more like the production figures of a shopfloor foreman. In fact, the latter would not be too far from the truth. He runs a virtual factory that is customizing the solution, testing code, fixing bugs, and training. He sits in meetings with SBI officers to set standard operating procedures and suggest any changes, if necessary. After a year and a half of hard toil, he is smiling today. "I have been in several large projects in 14 years at TCS but this has been the most challenging one," he says. What started as a trickle is now a roaring river, with dozens of branches going live every day.

Core banking gives tremendous power to the bank’s leadership to be nimble in the market but it also imposes a lot rigor on the employees. For example, how do you open an account? What information do you need to input? And who double checks each step and signs off? All this seems cumbersome initially, but the people who have used the system for three months and longer are getting the hang of it and enjoying the benefits.

“We are only training the trainers but State Bank’s HR is doing a wonderful job of evangelizing and letting the knowledge percolate through the bank, despite difficult condition,” says Ramanujam. When you are short staffed, it puts tremendous pressure on others if you have to send 25% of the employees for training by turns. But the staff has risen to occasion, and the benefits have ripple effect.

Despite an environment that was not friendly to computerization in the banking sector for three decades (1960s-80s), SBI has had a tradition of experimenting with new technology. According to Uday Shankar Roy, currently Chief General Manager of SBI, Kolkata who, till recently, headed the core banking project, SBI was one of the first banks globally to use the IBM 1401 mainframe for interoffice transactions an reconciliation, back in the early 1960s. "In the '70s, as load increased, we brought in brand new Burroughs machine, which was cutting edge at the time. Branch computerization started in the ‘80s and '90s, and a clear cut IT strategy was put in place in the late '90s. It started with local area network and branch automation, along with a massive interbranch networking project to connect over 4,000 branches. This laid the ground for taking up core banking two year ago," adds Roy.

Though the B@NCS solution chosen has been implemented elsewhere, the State Bank project involved a scale that was an order of magnitude larger and which demanded a high performance solution. "After all, the largest core banking project implemented in the world so far involves three different systems in different countries and a total of 2,400 branches. Here, it was 9,000 branches of SBI and 5,000 branches of its associate banks, all on one system. Scalability was a prime issue. It was only because we demonstrated it that TCS won, despite tough competition," says Jagdish Bhandari, head of the financial services practice at TCS.

Did they have to change the code significantly to customize it for State Bank's needs? “Of course. It is almost a new package now. The original package had about 3 million lines of code. Now it has 6 million!" says Ramanujam.

Why was such drastic change needed? What were the challenges? The original product, in its basic form, was a simple online transaction processing system, which was scalable. But, in this case, the customer made demands that created conflicts in the system. Basically, scalability depends on the product doing only one or two things in an efficient manner. Secondly, it has to adhere to some basic principles of databases. Scalability and maintainability require that all the information required should be stored at only one place. However if the system is expected to generate reports that involve data from different tables, this pulls down performance of the system. High performance requires normalized databases and puritan view of database design. But that would not meet branch requirements in SBI because such a system would not generate reports in the required format.

An online transaction processing system typically does not have a database designed to suit report generation. However, the branches felt that, unless reports were easily generated, their daily operations would not be complete. Yet even if the system did generate reports, it could still not handle 100 million accounts!

“We had a monumentally difficult time doing the correct tradeoffs,” says Ramanujam.

N chandrasekharan, EVP at TCS and Head, Global Operations adds, “The key to our success were the right design for the database and the program. It was a combination of and in depth knowledge of technology and banking business. We have demonstrated our competence to build large scale high performance systems.”

But at what cost? “Oh, we are building probably the lowest cost system in the world. In China the cost per branch for this technology is $25,000. In other developed countries, it has been $50,000-plus. But, in India, we spent only $11,000-$12,000 per branch. This is a very big advantage to us,” says Purwar.

How was that achieved? It is said that TCS reduced SBI's costs by not going for a mainframe based system. But Ramanujam disagrees. "Cost is not a function of just the hardware or infrastructure that you put in. The total cost involves hardware cost, software cost, maintenance and modification cost, and so on," says he.

"Just like a car. If it costs $11,500 but 'breaks down every 10 days, then we end up spending a lot more. We might as well buy a better car that costs more up front but needs less maintenance. Similarly, if software is poorly designed and cannot make the modifications required in a dynamic business environment, then it will prove expensive. For example, the Finance Minister may declare a new Cash Withdrawal Tax. But if the code cannot quickly absorb it, then it is poorly designed. Moreover, new product introductions should be very quick. In fact, recently, State Bank came up with four new product ideas in record time. One of them was rated as the quickest product that the bank had brought into the market ever.

Subramaniam Ramadorai, CEO & Managing Director, TS says, “The challenges in the implementation of a solution in a large banding group are manifold. It is due to great teamwork and hard work form the SBI and TCS teams that we have, to say, reached a stage of aggressive solution rollout across branches. SBI is truly a ‘bank that the nation banks on.’ We are privileged that it is banking on us for this unique project, which will tremendously impact the operations of the entire SBI group.”

All who doubted the capacity of this giant to change are now watching it transform itself rapidly. When complete, it would make for a great case study in change management for global managers. Clearly, in five years, India will have a new financial powerhouse.

Our aim is to provide ‘Class Banking for the Masses’, proclaims Purwar. That would indeed revolutionize banking in India as never before.
We say Amen.

Monday, September 21, 2009

Dr Keki Gharda: A profile


Eccentric Entrepreneur, a Genius—Dr Keki Gharda

We reproduce Dr Gharda's profile on the occasion of his 80th birthday, which is being celebrated by friends, colleagues and well wishers in the Indian chemical industry on 26th September, 2009

Excerpted from, “India's Technology Leaders” by Shivanand Kanavi, Business India, July 4-17, 1994

Perhaps, the incident that, best sums up Dr Keki Gharda, the leader of Gharda Chemicals, is a story that has become part of Indian chemical industry folklore. In the early 1980s, Dr Gharda was invited to an Monopolies and Restricted Trade Practices Commission’s hearing, which revolved around a multinational's application for a licence to produce isoproturon. Some Indian companies objected to the application for fear that they would not be able to compete. But Dr Gharda, alone, was of the opinion that there was nothing superior about the multinational's technology. To prove his point, he declared that his company would introduce a new, more efficient process within 18 months and compete with the multinational.

Most of the people who comprised the stunned audience that day would not have imagined that Gharda would be able to pull it off. But he did. True to his words, Gharda realised the danger of using the highly toxic isocyanate route in making pesticides, and came out with a process to produce isoproturon, using urea. Today, he is the second largest producer of isoproturon in the world with large exports to Europe, US and the Far East. The superiority of Gharda's process, which is today internationally known as the ‘Indian process’, made even Rhone Poulenc, the European giant which had a monopoly in isoproturon, sweat.

Since then, Gharda Chemicals has become the leading Indian company in technical grade agrochemicals. Cypermethrin - a popular insecticide; cypermethric acid chloride - an intermediate for cypermethrin; anilophos - a herbicide used, for rice, originally discovered by Hoechst though it now uses Gharda's process; chlorpyrifos- a new generation insecticide; and napropamide -- another herbicide, have all made Gharda a power to reckon with in the global agrochemical scene. In the last three years, a number of international agrochemical magazines have written about Gharda and even put him on their cover.

Gharda Chemcals started as a three-men-in- a-garage operation in the late 1960s, in Santacruz, a suburb of Bombay. Dr. Keki Gharda who had returned from the US after a PhD in Chemical Engineering from the University of Michigan, Ann Arbor, joined the faculty of University Department of Chemical Technology (UDCT), Bombay. However, he could not get a permanent teaching position and he turned instead to entrepreurship.

His first success, in the field of dyestuffs, came fast. He synthesized pthalogen blue, a dye that was very popular in the textile industry but had to be imported from Germany. Later azo-dyes provided the bread and butter. However, with the tremendous proliferation of process technology in the small-scale sector, dyestuff manufacturing was no longer attractive, so Gharda changed course midstream into agro-chemicals.

This time it was not as easy. The early 1980s, the years of transition, were difficult. As a plant engineer in Gharda’s Lote factory remarked, “In those days we were literally living hand to mouth. But unlike other industrialists, Dr. Gharda did not retrench any of the R&D staff. The later years vindicated his visionary faith in in-house R&D”.

That vision has helped him to move with the times. When quality and purity were absolutely essential to gain entry into export markets of Europe, Gharda welcomed the challenge. Today, his analytical lab is not only top class but has made a number of original contributions: half a dozen of them (in Fourier Transform Infra Red spectroscopy and in High Pressure Liquid Chromatography) have become part of international standards.

Today, Gharda is going through another round of diversification into bulk drugs and engineering plastics. But why get into bulk drugs when there are so many players already? “While the Indian' bulk drug industry is strong in organic synthesis, they are not so strong in chemical engineering,” answers Gharda. “With our strength in both, we will be more efficient.” In their typical style, Gharda Chemicals are building plants for products that are yet to come out of their R&D.

But this confidence has got him his fair share of detractors. “He is an eccentric,” says an indignant major player in bulk drug manufacture. “He wants to spoil the bulk drug market by driving down the prices.” His anger is partly fuelled by the not-so-polite letter he received from Dr. Gharda, which categorically stated that since Gharda Chemicals is going to enter the bulk drugs market in fluroquinolones at a lower price, the existing competitors might as well quit!

Therein lies the essence of the man. Variously described in the industry as a missionary, a Gandhian, an eccentric, and a spoilsport, he has nevertheless always approached the market in his own unique manner. His strategy is to come out with new products at prices that are at least 20-30 per cent lower than the prevailing ones and hold them at that level for years. Aside from driving the competition out of the market, it even makes giant multinationals wary of him, perhaps one of the reasons why some of them are queuing up to tie up with him.

And he lives by his own rules. For examples, though he has received many awards for novel processes, he has not patented any. He believes that the superiority of technology should be decided in the marketplace rather than in court rooms. He also shies away from breaking up the manufacturing process into a number of stages and carrying these out in different plants to guard his trade secrets, as most of his competitors do. “This will lead to the right hand not knowing what the left is doing. And you would lose the team spirit where you troubleshoot together and learn from each other” he says. “The name of the game is to come up with newer and better processes and products through R&D and always stay a step ahead of the competition.”

Friday, September 4, 2009

Rajiv Motwani Interview Part-2


Rajeev Motwani Unplugged-II
(This interview was conducted by Shivanand Kanavi at Stanford University, Palo Alto, CA in July 2002)
For the first part of the interview


Rajeev Motwani: The last one year I am spending a lot of time at Google, at least an hour or two every day. In the summer I am spending half my time at Google working on research projects and the next generation. These guys are working on something that I could contribute to and I am happy about it.

Shivanand Kanavi: What is the next generation? People are talking about contextual searches, semantics and all that. Can it be done at all, it all sounds like a wish list.
RM: Science progresses by making up wish lists. They are always unattainable but on a scale of one to hundred if one takes three small steps it is a big thing. How get appearance of semantics and context, without solving the big problems of artificial intelligence.

SK: Machine translation itself is big issue.
RM: Yes but a lot that one can do before you get there which is in the direction of context and semantics. Google has 50 plus PhDs among about 300 people. A lot of them good Stanford PhDs a lot of them do machine learning, artificial intelligence, systems, algorithms… everything. I am enjoying it.

SK: Is there any competition for Google?
RM: There is always competition. Once when I was interviewed by TIME magazine, when Google was just being set up as a company and the PR people at the Google said it would be good if you talk to the journalists. I said Ok but I did not know that you have to talk to them differently. You have to be careful with them. In the end while they were packing up they said by the way is there any competition to Google? I said ‘the biggest competition to Google is Google itself. There is always arrogance of youth. One forgets that just as they were 21 year old at one time when 30-40 search engines existed and they started with a shoe string budget. (I know it because I was one of the contributors.) With less than $50,000 dollars they started this, which has become so big. They did not know that they have no right to do what they were doing and succeed.’ They put all of that in the article in TIME.

SK: That is anyway good advice.
RM: The problem with being young, I was there once, is that you always believe that you can do whatever that you set out to do. Usually you are wrong but some people turn out to be right and go on to change the world. Someone will figure out a better way of doing things than us. That is why Google is hiring all these smart PhDs. They are doing a good job so far. But I will never underestimate the PhDs that are coming out. But you need to be very smart to do better because it is a non trivial thing to do better than Google.

SK: If the next step is understanding the question and then searching for the answer then the machine should be able to first understand the question. When we don’t have that how are we going to go ahead. Then there are people who are talking about an Indian language Google!
RM: Even what Google is doing now, which is very shallow semantics, in Indian languages is a very big thing. Without trying to build it from scratch can you do in other languages what you have done in English, that itself is a big challenge and there enough people in Google working on that. It is hard for various other reasons also. Google claims it has 2 billion pages. I don’t know how many they have, but let us say they have 20 billion pages. Each page has 1000 words and if you are getting a billion queries a day, then you have to search through a trillion words a billion times a day. Can you imagine the scale of that! You might have the best technology that understands the semantics and context and all that but how do you build a machine that does it billion times trillion every day! I have seen other people who have shown to me that they can come up with better answers than Google if they are given 100 web pages. If they are given 10,000 pages they take an hour and if they get a billion then they are never going to do it. So the greatness of Google is not just technology, understanding the structure etc but doing it in 0.2 sec regardless of the query.

SK: Does the answer lie in quantum search algorithms?
RM: That is still science fiction. Even if you were able to make them work, making tens of thousands of machines work with each other is a different ball game. There are pure system problems and not AI or search algorithms. Storing, indexing, searching and then when thousands of queries come each second you have to make them all happy by giving the answer in 0.2 seconds. If you take 0.5 seconds then they are not happy. They are not paying for it but they are unhappy then they may not come back and what keeps Google’s business model going is that millions keep coming back.

SK: It is a great project no doubt. People are talking about building Universal Digital Library. But if you have scanned and built a large digital library finally you need to search out the relevant information. It is back to Library Science! The digital library is supposed to be one great spin off of the Internet technology.
RM: Yes. I used to go to Stanford library on the average twice a day. I have not set my foot in the library since ’96. Everything I do sitting on my machine here. That is the difference it has done to my life. Everything is on the web, unless it is some paper written 50 years ago and is only in hard copy. If that is the case then I should go and do something else anyway.

SK: Besides Google what is provoking you intellectually?
RM: I get bored too soon in everything I do in life. The new thing in my life is the start-up work partly motivated by what I saw happening in Google. I invest in companies, I mentor companies, I sit on board, the whole business side of it. That has its own challenges. It stretches different parts of your brain. It is a strange mix of common sense and technology. I have invested in software companies, box companies, security, chip companies, storage and search everything. I have become a start-up junkie right now.

Monday, August 31, 2009

Talk: India and the Digital Revolution

I delivered a talk on "India and the Digital Revolution" at the Institution of Engineers (India) Dharwad Centre on Aug 21, 2009 and greatly enjoyed interacting with the audience. Below are a couple of reports that appeared in the press and blogs......

http://timesofindia.indiatimes.com/NEWS/City/Hubli/Taking-a-peek-into-digital-field/articleshow/4946374.cms

Taking a peek into digital field
TNN 29 August 2009,
DHARWAD: India contributed significantly to the digital revolution in the 90s, though it did miss electronic revolution in the 60s, said Shivanand Kanavi, vice-president (special projects), Tata Consultancy Services (Mumbai).
Delivering a special lecture at the Institution of Engineers' local centre here, he narrated the comforts achieved because of digital revolution in the fields of industry, trade, transportation, education, banking sectors, administration, medicine and communication, and others.
"The digital revolution provided global platform for research and development," he said, and highlighted its impact on society through media and entertainment. He made remarks about the contributions of Indians to the digital field.
The lecture was followed by an interaction with audience in which questions of common concern like opportunities for youngsters, future scope in chip technology, etc, were answered.
Local centre chairman Mahesh Hiremath, Prof. Ramesh Chakrasali and honorary secretary prof. Mrityunjaya Kappali were present.



http://reportingweb.blogspot.com
Saturday, August 22, 2009
India and Digital Revolution: A Talk by Mr. Shivanand Kanavi, VP, Special projects, TCS
The Institution of Engineers, India ( IEI), Dharwad Local chapter organized a talk on "India and Digital Revolution" by Mr. Shivanand Kanavi, Vice President, Special Projects, TCS. Mumbai, on 21st August, 2009 at Balekundary, Hall. Mr. Kanavi has an illustrious career which took him to different domains such as research at IIT, teaching, Business Journalism at Business India. He joined TCS in 2004.
The topic of talk " India and Digital Revolution" was a very thoughtful choice and Kanavi did a great justice to it. His talk composed of two parts- first dealing with the technological underpinnings of digital revolution and its impact on India and second part was about Indians who contributed heavily in shaping digital revolution. The impact of digital revolution in India was termed " fall out" by the presenter and he listed the following as the main areas that benefited Indian- people, market, business.
Indian IT
Chip Design
Telecommunication
Global Platform for R & D
Centre for Engineering Design
Media and entertainment
Governance
The root of all positive impact of digital revolution is due to advancements in telecommunication. The developments made it possible to separate design and fabrication functions and get them done anywhere in the world. The Indian talent took this opportunity very well and now "India strategy" is central theme to any major business in the west.
The chip technology has become ubiquitous in its spread and virtually every tool that we use today contains a number of special purpose chips. For instance, a modern day car has a number of chips for steering control, wheel etc. Companies such as Texas Instruments ( TI) are doing a lot of R & D work and chip design in India a lot of cutting edge work by GE and others is being done in aerospace technologies in India and so on.
Historically India had missed earlier revolutions - Industrial ( due to British rule ) and electronic revolution - however, the telecommunication industry made us to catch up and join the revolution in later stage. 80% of the mobile sets used world wide use DSP technology designed by TI. GE has established a R & D centre in Bangalore that contributes upto 40% research work in aircraft engine design.
The design and development of Ferrari used in Formula -1 is designed by TCS. Tata's Nano project proved that with just one fourth of research budget allocated in the west, one can design and deliver a car from concept to market.The contribution of ISRO towards making India a strong player in world is very great.Starting from its SITE program for distance education in 70's, ISRO has technologically supported Indian march including introduction of nation wide colour broadcasting during ASIAN games in 1982.
The devlopments in digital technology have made a great contribution in changing the way government runs it services. The land record computerization has resulted easing of many services. The use of IT in central government departments such as company affairs has enabled them to handle data of more that 6,00,000 companies very efficiently and in a timely manner. The IT use in passport issuance has greatly simplied the very process and it will be possible to get passport within three days of police clearance.
Similar changes have been planned for judiciary, medical departments. Digital library is planned as solution to reach wider learning audience.
Implementing DEMAT mode of shares and computerized transcations in stock market is another example of IT in action. The NSE in India is one of the busiest stock exchange in the world and its systems have been built using economic hardware and software components. The Role of IT in banking is still more stupendous- SBI with more than 14,000 branches has been successfully using IT to serve its customers in every nook and corner of the nation.
The Indian railways took the path of computerization without resorting to massive lay off in the ‘80s and its success has brought a sea change in the public perception of computerization. In summary IT has been playing a pivotal role in Indian current development context.
The second part of the talk was profiles of some perosnalities that contributed in the making of digital revolution. The list included J C Bose, Amar Bose of Bose System fame, Raj Reddy, Praveen Chaudhary, Arun Netravali, F. C Kohli, Abhay Bhushan and many more.
Shivanand Kanavi's talk was one of the most informative and inspiring talk that I had heard in the recent past. Thanks Sir

Saturday, August 29, 2009

Rajeev Motwani: Interview 2002


Rajeev Motwani unplugged


(Excerpts appeared in India Abroad, Aug 7, 2009)


Shivanand Kanavi, interviewed Rajeev Motwani in July 2002, while researching for his book, Sand to Silicon: The amazing story of digital technology. Here are the edited excerpts:


SK: Tell me about your childhood and growing up and the influences that shaped you.

RM: One of the shaping influences was that my father was in the Army and that meant not being in one place for too long, not more than 2 years. My parents were great believers in education so where ever we were they sent me to the best possible school available. All of them were missionary schools and had many inspiring teachers. I also always wanted to be a mathematician or a scientist at any rate. Then I decided that I did not want to be an Einstein but wanted to be a Gauss. That was because I was an avid reader and I used to read a lot of books. My parents had given me a lot of ten great scientists, 5 great mathematicians kind of popular science books and biographies which were very inspiring. I was not reading about other kind of heroes. That is what I wanted to become. Fortunately I was good at Maths. I also did not graduate from school. I used to study in St Columbus, Delhi, where they had just switched from 11 to 10+2 and IITs gave us permission to join them after 11th without completing 10+2. It was just for that year. I did have problems when getting permanent residence here because I had a PhD but did not have a school certificate.
IIT Kanpur at that time had for the first time an undergraduate program in computer science B.Tech. I really wanted to be a mathematician and I did not have any idea what a computer was. My parents were hesitant because they did not know how a mathematician would make money and support a family. Basically I was forced to do Computer Science. I then realized that Computer Science was very closely related to mathematics. Some of the faculty in IIT Kanpur were also a shaping influence for me. One of the people who really influenced me was Kesav Nori. At that time there was Prof Rajaraman, R.Shankar, Sahasrabuddhe, Somnath Biswas, Kesav Nori, Harish Karnik to name a few. I could not have constructed a better environment for doing computer science in India. It was an amazing confluence of people.

SK: But they already had a masters programme for a long time.

RM: Nori had just come back from Europe. He stayed for a year or so and taught the first course in programming. He was a wonderful teacher and used to tell great stories. We started out programming on comp cards, which you probably remember but most other people don’t. That time we used to work on DEC machines and Vac machines with a terminal. We then had to use a login and a password. Nori could have made up random passwords, or give names of flowers but instead he gave names of famous computer scientists as passwords. Somebody had Don Knuth as password (who is down the hall).
I went and did research to see who these guys were. Bob Floyd was my password. He was also at Stanford and passed away recently. He was one of the mentors of this field called analysis of algorithms. He also did the early work on randomized algorithms. I ended up eventually did some work on randomized algorithms. The very first chapter of my book on randomized algorithms was on Floyd’s algorithm. It is hard to believe that because he was my password this happened. But there must have been some connection! That was the wonderful thing about Nori who was a very inspiring person. He did more than just teaching. He created such a wonderful ecosystem and developed a personal connection with his students.

There are a lot of very good schools not only in the US but elsewhere. Going by what I have seen in this country and going by what I have learnt out there, definitely IIT Kanpur was one of the top five schools in computer science education.
I finished all that. Everybody else was coming to the US for PhD or Masters or whatever. Actually I did not want to come here for reasons I don’t quite understand now. I remember getting a job at DCM Data Products because getting visas at that time (1983) was a big problem. I was then interviewed at Wipro by the top three guys. It was a small outfit then.

SK: Was it Ashok Narasimhan?

RM: It could have been. I met Ashok Narasimhan last year when he was doing a company called July Systems which I was considering investing in, but he did not remember me. The interviewer said we would love to give you a job looking at your track record but isn’t every one with your kind of back ground going to the US on a scholarship? So have you applied to US? I said yes I have an offer from Berkley. He asked do you have a scholarship? I said yes but I am not sure if I will get a visa. He said I did my MBA from Stanford and believe me you would want to go there.
If you come back we will give you a job.

Berkley was very different from Stanford. It was a very politically oriented university. You could call it the JNU of the US because it was highly politically charged. Ronald Reagan was the president then. So then for 3 years I had a blast. Did not do any work and fully enjoyed the environment. My advisor was Richard Carp, who won the Turing award – which is like the Nobel Prize in computer science in 1985-86, when I had finished those 3 years. It was then that I thought that I was not doing anything and letting this man down. So from then on I worked really hard and was quite productive for the next two years.


SK: What did you work on?


RM: My PhD thesis was on randomized or probabilistic analysis of problems in optimization. Problems in network flows, graph matchings and so on. These are general formulations of a large class of problems.

SK: Traffic problems?

RM: Yes traffic problems or network routing. Routers on network, are basically implementing matching algorithms in some form, that is at the micro level. At the macro level flow of packets on networks. My advisor, Dick Carp was a pioneer in that field. These problems were hard and so I was trying to find a heuristic faster and better solution to get the right optimal value. So that was what my thesis on.

SK: People in the telecom sector had used such things.

RM: The technique itself had been used to some degree. In fact pioneered by Carp. Called NP- completeness. It says that some problems are essentially impossible to solve. They cannot be solved exactly. So the question is if you can solve the problem approximately with some assumptions such that the instance of the problem or the input is randomly distributed with known distribution. That’s what I was trying to do.

SK: Why random? Connection with Gauss again?

RM: Well the reason being that once you assume that there is distribution, then you can give some structure to the problem and you can use some probabilistic techniques to say that on a certain fraction of the inputs I’m still going to get screwed up but if I bound the fraction on which I wont perform well and with a typical example I will get a good solution. This is a problem with the theory of algorithms as it exists. In those days everything was taken from the worst case point of view which means that if I want to route a flow on a network I can always construct an example where you are unable to route the flow and perform badly. But the real flows patterns that emerge are not the worst case patterns. They have some niceness to it and things do get routed. Randomness is a way of capturing that by saying that in distribution there is a probability that you will get bad flows but many times you will also get good flows and that is good enough.

So I was doing all this and was about to graduate and was wondering what to do next. To go back to India or stay in the US because again other people made the decisions for me, Don Knuth, one of the founding fathers of comp science, also one of Nori’s passwords, came over to meet my advisor and told him that they wanted to hire someone young for algorithms at Stanford. So Carp suggested my name. I was then invited by Knuth at Stanford for lunch during a dinner hosted for him at Berkley. I was wondering why this great man wants to have lunch with me. So I went to Stanford and met him at a restaurant near the church at the quad. He then told me to be with Stanford for a year and see if they liked me and vice versa after which if things worked out well they would hire me.
For the first year I was a visiting faculty. I did not want that job as I was getting better offers and permanent jobs at other places but since it was an offer by Knuth it was hard to turn down. It’s the same as Einstein inviting you to Princeton for a job. So then I came and taught at Stanford and started some courses and had a very good time.
I was then given a permanent offer at Stanford as they liked what they saw. It was 3 – 4 months after I finished my 1 year at Stanford that I was to get married. My wife was shifting from LA to Berkley. Then we saw that Stanford was a good place and decided to stay in Stanford.

SK: What have been your major research interests here at Stanford?

RM: Teaching has been a major preoccupation first of all. I enjoy teaching. The reason I was in Stanford was that there were many faculty retiring like Don Knuth and so they needed someone to fill in and walk in their footsteps. So since so many people were retiring or leaving there was a vacuum created here. Also there were a lot of courses to be taught. I ended up teaching all these courses. I even made my own courses like topography and algorithms and complexity theory. I did not get enough sleep as I did not know a lot of these areas but I did learn a lot through teaching these things. I am a perfectionist and still get nervous to talk before a class till today. I get nervous, what if someone asks me a question and I find myself unable to answer it. So for this reason I always over prepare.
So this nervousness has taught me more than what I learnt as a student. I now have worked in many different areas and it broadened me up. I have the tendency of getting bored very easily and so if I stay in one area too long I quickly move over to another area. My threshold of working in one particular area is about 5 years.
Some of the non obvious areas are robotics. I was inspired by Jean Claude Latombe from France who was in this Dept. He told me that there were a lot of algorithms in robotics which are needed to plan the actions of the robot. Robots in a generic sense could include cars and or any mechanism which has to plan a motion to move about. It’s like the human body wanting to get up from this chair and walk to the door. It may seem like a triggered action but there are a lot of complexities and degree of freedom involved. In the human body itself every joint in the body gives a degree of freedom. Each can be controlled independently by setting the angle of each joint to accomplish the task. The reason why humans have this degree of freedom is for them to operate in the real world and do a lot of things. The control of these degrees of motions becomes very high. Although we live in a three dimensional world, the robotic movements and freedom work in a higher dimensional surrounding. So if you send a space craft to Mars, then you have to be sure by planning intelligence that it goes the right way without hitting anything along the path. This requires very high dimensional planning. It is like having a starting point A and end point B in space, and moving from A to B without being hit by any obstacles. The same task would be easier with 2 points on the table. So the space that we are talking about is not the physical space but the space of complex possible motions.

SK: There are constraint surfaces?

RM: They become very complex constraint surfaces in high dimensional geometry. I learnt this space for a few months and realized that this problem could be solved through randomization. It is very hard to plan motion in high dimensional complex places while it is very easy to pick a random point in space and figure out if it is going to hit any obstacle in space or if it is a free part of space.
If you pick many random points it is very easy to sample but very hard to find a free point in that space. If you find many free points, then you hook them together and make a path. The path may not be the smoothest but you can smooth it later. So that was the fundamental idea we used. But to realize this and analyse it, apply it and turn it into real systems is a lot of work and I worked for 5 years on this space, putting high dimension geometry and randomization together. Jean Claude was a systems guy while I was the theoretical guy. So the students implemented some of these things and these were used at places like GM in their robot assembly lines.
None of these things were used in isolation. The actual robots do many things and one of the techniques they use is this. They have many softwares running them. That was the first time I did something mathematical but practical. I suddenly realised that I could operate in the real world after seeing the robots move and perform using my algorithms. I could now move over from the paper pencil world. I credit Stanford for creating an environment where people in different areas can work together – the whole is greater than the sum of its parts. So I spend one third of my time in doing my work in a theoretical and mathematical way and then I collaborate with people. I may collaborate with someone in networking or databases or compilers and use my skills to solve their problems. But I need them to use my mathematical problems to solve the real world problems. It has been a good synergy so far.
I got the Godel prize for my theoretical work. It was a very theoretical work. In science it is said that one guy stands on the shoulders of another and another on his and so on. The guy on top gets the prize. In my case I was on the tip of the pyramid and so got the prize. Everyone forgets the pyramid. In my case we had worked and come up with excellent results before my paper which we used and leveraged to prove what we proved. We proved that you take a mathematical theorem then there is a way of writing the proof of that theorem and there is a way to easily verify for correctness. Verify correctness means for eg - Take the theorem and the proof and probe the proof in 7 places, read what is written there and then give an answer if the proof is correct or incorrect. The answer would be correct to as high a probability as you want. Random sampling of 7 places. Irrespective of the length of the proof. They would work for any given proof provided it is in a specific format and a particular mathematical language.

SK: What makes you say that the whole is correct?

RM: That is probably too hard to explain. The basic idea is that of an error correcting code. The code takes a sentence or a sequence of bits, 0’s and 1’s and replaces it by a slightly longer string of 0’s and 1’s , but there is some redundancy built into it.

SK: It is like writing a checksum.

RM: It is exactly like writing a checksum. I am giving a very simple example.
There are ways of error correcting where you rewrite small bits of the whole thing. It may not even look like the original thing but from that you can extract the original message. So take a proof of a mathematical theorem and write it in an error correcting code. If there are a large no of errors then by random sampling of a few points in that I will detect that there a large no of errors. So I will dismiss the group as nonsense if there are a large no of errors. But if there are small no of errors then I can correct it because I know there is a correct version of this. So after sampling it if I find the number of errors to be small then I know that there is a good proof somewhere.

SK: So it is an existence theorem?

RM: It is random verifier. We over simplified it in this discussion, however the implications are important. Sometimes in mathematics you do things just for the sheer elegance of it and after we wrote this paper I learnt that Intel had a problem with pentium2. When you multiply 2 specific numbers on that p2 chip which was being used in all the desktops and laptops, the answer was wrong. An overflow perhaps. I got a call from Intel to ask me if there was any way they could use the verification technology. But it was not possible as I was a purely mathematical abstract trained whereas they needed a real system with 10 million gates on it where you had to do some checking. Very hard to translate it to this. But this did show the possibility of verifying errors in a complex system by doing a small amount of work. But it required the system to be written in a certain code which was the catch.
There is a wonderful mathematical theorem, I am quite proud of it, it is a combined effort of a lot of people in our field, and is one of the deeper pieces of mathematics which has been done in modern day computer science.
It turned out that implications of this are much more profound than the statement itself. It goes back to the motion of optimization problems we were discussing earlier where we were talking about the flow algorithms where we couldn’t get an exact answer because you can show that in a “polynomial” amount of time which is a measure of efficiency in computer science, one could not get an the exact answer. In my thesis I had looked at one particular way of getting around the problem which was to assume that the problem input itself was randomly distributed. And I was not faced with the bad cases alone but was faced with a distribution of cases; some good some bad. But on the average I would do ok. There is also some other way of tackling hard optimization problems which is to say that I still guarantee you the right answer but the right answer won’t be the optimal answer. So if you are trying to route the maximum no of circuits through your network, I will not route 100% routing but will route 95% and to get 100% is an impossible computation. So these are called “approximation algorithms”.
So for some large class of problems we faced, using this theorem we had the possibility of reaching an approximate answer but for some problems reaching the approximate answer was also not possible. That was one of the bigger breakthroughs in comp science. Carp and Cook from Berkley in 1971 came up with the theory of NP-completeness, which tries to describe why some problems can only be solved by reaching the approximate answer and not the optimum answer. Twenty years later as a consequence of this we said that for large subset of their problems not only is it not possible to get the optimal answer but also hard to get the approximate answer which means that there is no use trying to even solve these problems.
In science you have the family tree where people are followed by their advisors, juniors or siblings. Where each person follows up on another’s work thereby expanding it. Like the musical schools in India, where the Gharanas pass on the knowledge from one generation to the next. I have been a beneficiary in that kind of a system. To me it all came together very nicely. No one knows what happened underneath to build it up. A lot of people contributed to it. I was at the tip, a small part of it, towards the end point.
That has been my mathematical pinnacle. After that it has been downhill all the way (laughs). So I did this and robotics and compiler optimization. I did PLIW compiler optimization and then Pfizer wanted to fund research on computational drug design. And while finishing the work on random motion planning in robots we realised that molecules and robots actually behave in a very similar way.
What is a molecule? One model of molecule which we learn in chemistry in high school is atoms are like balls bound together by bonds thereby making a molecule. So for eg. a protein molecule which has about 30, 50 thousand atoms has to fold and form a shape for another protein molecule to come and bind on that fold. The place where it binds has the activity.

SK: They are called pharmacophores.

RM: Yes. This is basically drug design. It is all about figures folding and matching like a lock and key mechanism. We said we know how things fold, we know how degrees of freedom are created in high dimensional space. Let’s throw it at this problem. So Pfizer funded this research which went on for 2 – 3 years and we came up with software based on our theory and we added some new theory. On a particular approach, the project being called RAPID (Randomized Pharmacophore Identification for Drug Design). It went very well. I learnt a lot. It was an intriguing experience. I had to go back and learn my high school chemistry and biology and the other fun stuff. And the software that we made is being used by Pfizer labs for their drug design. Of course the problem is complex and needs more than one software. So due to confidentiality issues they didn’t tell us how our software was used or what drugs were made from it but you feel you have made a contribution to society instead of just doing formal mathematics which is hard to justify. It felt like we contributed to social welfare through mathematics.

SK: These techniques are mostly used by medicinal chemists. They have an intuition as to which molecules to use and how the binding would take place.

RM: Now they have a tool wherein they can enhance their intuition, do away with some things and focus on what they really want, which is what we allow them to do.

SK: It can be applied to catalysis and designing chemzymes as well right?

RM: Yes. We never got into that because by then the world wide web was coming up and I just got sucked into that. There was this guy Jeff Ullman, another one of the grand old men of computer science, who retired this year. He was in the office next to me and was in database. I was talking to him and a new student – Sergey Brin, and I remember at that time we were using Mosaic, and we were looking at the web and I was sitting there and thinking that we could randomize the web in some way because that was going to grow and become big and randomness was going to be important; though I did not know how and why. So I thought about doing random walks on the web and there was this problem of crawling on the web. At that time a search engine called Inktomi had just come out of Berkley. Excite and Yahoo had come out from Stanford so we had seen the first signs of all of this.
I remember going to Inktomi and searching for the word Inktomi and it could not find itself. I don’t know if that is still true but at that time if you went to Inktomi and typed in the word it said no results found. My Godelian past induced me to do these self referential queries but what amazed me was that this is a simple thing that people screw up on. So in the context of all this I was listening to some people from IBM talk on Data mining and Ullman had just introduced me to some problems in databases. I broke them down with a student and was getting pretty excited about the concept of databases. Ullman took me for this talk on data mining which sounded very interesting to me. So Sergey and Ullman and we decided to do some data mining on the web because it sounded like a nice mix. We then formed this research group called Midas which stood for Mining Data At Stanford. We did a lot of good work on data mining. Then there was this guy called Larry Page who wasn’t really a part of the Midas group but was a friend of Sergey and would show up for these meetings. He was working on this very cool idea of doing random walks on the web.
When I understood what the World Wide Web would look like, I knew I had to somehow force randomness into it. When Larry showed us what he was doing, it was like a complete epiphany, we thought it was absolutely the right thing to do. So Sergey got involved and it became a sub group inside Midas. I was really a good sounding board for Sergey and Larry and I could relate to what they were doing through randomness. They then created a search engine called Backrub. It was running as a search engine from Stanford just like Yahoo ran till the traffic got big and the IT guys sent it off the campus. So these 2 guys would come to the office and say “hey we need some more disc space”. They were completely non respectful of me, which was a wonderful thing. They treated me like an equal. These 21 year old guys were demanding things from me. They needed more disc space because it’s getting bigger. So we need more disc and more money. There are still pictures around the building of how they used to use Legos, to create a box inside which the discs were being put. These discs were those cheap ones bought from the back of a truck and were generating a lot of heat. So they put it in Legos to allow for air circulation.
For me it was a fun research project. We had a lot of ideas which we shared. At some point this thing started getting very serious and we wanted a better name for this than Backrub. So somebody came up with the name Google. Google means 10 raised to the power of 100. It is actually spelt as GOOGOL but somebody miss spelt it and that’s how the search engine got its name. Of course the official story is we deliberately spelt it that way but my guess is we miss spelt it.
So Google started and pretty soon everybody in the world was using Google. The results were much better than all the other search engines going around. It was by word of mouth like I tell my brother to use it, he would tell his wife, the wife would tell her kids and so on. At some point these guys said we want to start a company. Everybody said it was not worth it. There were 37 search engines already there. How would you raise money? How would you form the company? But they decided to do it and they did it. There were some big names which supported the company. Andy Bechtolsheim, an ex Stanford guy who along with Vinod Khosla had founded the Sun Microsystems, put in a little bit of money. They managed to raise a million dollars. They started the company and it was right here in the university avenue. It used to be on my drive home so I used to go and hang out with these guys. It used to be wonderful.

Then they took over the world!

Right now the other search engines don’t even compare and I remember people who I don’t want to name saying why do you need another search engine? Today it is the only search engine people use. Now it is a company of 500 people or more doing hundreds of millions of searches every day, generating revenue. One of the few companies which in today’s economic conditions is not only surviving but also growing. Feels like I was part of a little bit of history and contributed to that history.

SK: Can you explain in simple words, the concept of search engines? How has it evolved?

RM: There was this thing called Information Retrieval which was used for document management. Consider your desktop machine. So let’s say you are writing chapters of a book and chapters of others books. And there are thousands of documents. Each document is just a document of English words. This could be more related to encyclopedias and newspapers articles and Reuters news feeds and so on. So you have all these documents with you and somebody comes to you and asks you a question. Can you answer that question or at least point to the right document that answers that question?
All of this was unattainable. So it reduced down to can you at least show me the document that contains the same words that are there in the question? If I said, “what kind of yeast should I use to make bread?” Ideally you would like it to come back with the answer. But you can’t do that. You could though at least find the document that uses the word bread and yeast and hope that has something to do with the question. That was the field of information retrieval. When the web evolved it became pretty clear that there was going to be a tremendous information explosion.


SK: Initially you had to give some key words for every web page and it would search only those keywords?


RM:
Even now it is still the same. It goes through the entire document and indexes the entire document. The computer helps to solve the scale of the problem which is huge. There are billions of web pages, each having thousands of words on the average. So there are trillions of words. So what you do is create an index that is trillion big and when you come along and say yeast and bread you find all the documents on the web that contain yeast and bread. It turns out millions of pages containing yeast and bread which is of no use to you. So people come up with this rule where if the document contains the word yeast 7 times and the word bread 3 times, then that is a score of ten, so give me the highest score document. I am just simplifying the rules of information retrieval but they have even more complex ones. For eg the word bread is very common and can be found in many places but the word yeast is uncommon so you start doing weighting and the other heuristics. None of this required rocket science, none of this is very deep.

It was library science. Not computer or mathematical science. So the old search engines like Alta Vista, Inktomi and so on, to a large extent this is what they are doing today too. The problem they were solving, the key secret sauce was the scale. They were able to harness a lot of memory and solve these problems on a large scale. So instead of building indexes of thousands of words they could build indexes of trillions of words and search through it very quickly. They just put a lot of horse power behind it but not any science. There were a few small ones which were exceptions but this is generally what search engines are doing today.
What Google changed was the following
It actually started looking at the structure of the web. The basic structure we noticed was that it’s a collection of documents. These documents talk to each another. The reason is the hyperlink. Without the hyperlink, the web would have been useless. When you click on it, it takes you to another document. It is like this document saying “hey, look at that other document”. Now if I create a web page and make the effort to point to your webpage, there is some meaning and connection between my webpage and your webpage. The content here and there should be related in some form. This was the basic insight in Google. Instead of looking inside the webpage or the words inside it, look at how the web pages talk to each other, how they talk about each other how they point to each other. So that was the basic insight and everything else was built around that insight.
One of the key things they did was coming up with this ranking function. So if you went and queried on yeast and bread what I would like to do is go to the most authoritative page on the web which talks about yeast and bread. So if there is a yeast or bread makers association of America, then presumably theirs is the most authoritative page on how to make bread out of yeast.
You do use yeast in a bread right? Don’t want to get that wrong. (laughs)
So the question was how to find the most authoritative page on a certain topic? Here is a simple basic idea that underlies that notion. The notion is – Look at the structure of the web. Secondly how to we convert this structure to a ranking scale? This goes back to the random walking I was talking about earlier. Suppose you are at my web page. There are 5 links out of my web page. You are a surfer. What did you do as a surfer. When we first surfed the web, we went to a page, clicked on it and found it magical. That clicking led to another page, and so I look at that and click and go to another page. And so now you are surfing. Suppose you are surfing at random. Let us say my page has 7 links. You randomly clicked on one of those 7 links. You reached the next page which had 3 links and clicked on one of those at random and it took you somewhere and so forth. After a while you will be on a random page on the web. Suppose you surf long enough, you do it a million or billion times, you will be distributed somewhere on the web. You could be on any page. The question is what is the probability that you are sitting on a particular webpage?


SK: It’s a graph theoretic problem.

RM: Yes it is. This is called doing your random walk on the graph which excited me about all these things. Turns out the probability distribution is not unique. There is a different probability being on different pages. Quite obviously if every page in the world points out to my web page then the chances of ending up on my page are very high because wherever you are there is some probability that you will come here. Nobody points to me or if one guy points to me then it is very unlikely. On the other hand if the important pages in the world point to me, then you are likely to end up at my page. But what are important pages? Those are the pages to which other important pages point. So this logic of circularity or flow is what led to the notion of page rank. Ranking of pages is Google’s secret sauce. So they came up with those and discovered that this is the right thing to do.

In hindsight I came up with the estimation of the random web surfer that there was a purely mathematical thing of eigenvectors and matrices etc. Now on a query on yeast and bread, what we do is look at all pages that contain yeast and bread , find the page with the highest rank or score which has the word yeast and bread and we say that must be your answer and it is usually right. Google got so cocky on this, that it has ‘I’m feeling lucky’ button. So if you give a query and hit that button, it takes you to a page. That page has the right answer.

Wednesday, August 12, 2009

Book Review: Curfewed Night

Peerless
A reaction to "Curfewed Night" By Basharat Peer, (Random House India, 2009)
Shivanand Kanavi
(See: http://wearethebest.wordpress.com/2009/08/10/an-example-to-emulate-for-indian-journalists/)

Basharat Peer’s “Curfewed Night” is a welcome first-person account of Kashmir of the last two decades. Peer’s book is lyrical, intense, partisan and cynical in varied proportions at the same time.

A simple linear narrative of events since the 1980s as seen by a Kashmiri boy (the author), Curfewed Night will help in educating the vast mass of Indian people who are distant from Kashmir in every way, who are not activists of the human rights movement, and who are the chief target of the Indian State’s one-sided propaganda about what’s been happening in Kashmir in the last two decades.

The book begins at the beginning that is the author’s childhood. This part is lyrical and at times cute. It could have been the retold story of any articulate, sensitive boy from any Indian village to any urban or exotic audience. Then comes teenage and the romance of the Azadi movement; the blind fury and brutality of the security forces clearly reflecting their hate and an occupationist attitude towards the Kashmiris.

Peer tells the story of the emergence of the struggle of Kashmiri youth, armed and trained across the Line of Control (LoC) by our friendly neighbours and the impact of all this on their friends and families. The author’s own brief inner turmoil to cross or not to cross the LoC, the romance of a sexy AK-47, and the pressure from the family to follow a more traditional middle-class road and, above all, a concern for self-preservation, are all conveyed very convincingly.

Then comes the life of a self-exiled student and later of a young journalist in the 1990s, with a longing to tell the “untold story of Kashmir”; the evolution of the author with exposure to a normal life and ‘freedom from searches’; exhaustion setting in about indigenous militancy with no hope of a quick victory and so on, seems a little rushed.

Peer then gives us an invaluable, authentic picture of the emergence of jihadis from Pakistan equipped with laptops and satellite phones ready to unleash terror, where the random victims are not necessarily military targets, while a hapless population caught in the cross fire continues to grieve over the loss of a generation.

Peer excels when he brings out journalistic gems like the story of the ikhwanis, turncoat militants who became a part of Indian counter-insurgency; chameleon-like careerists who smoothly switch roles between militant, reformed militant and politician, a cryptic hint of the alienation of separatist politicians from the ordinary aggrieved Kashmiris; or the schizophrenia of a swaggering para-military officer who unexpectedly melts in a media room when Peer starts recalling the life he spent in Delhi.

Despite these excellent points, however, there are some rough edges and glaring lacunae as well.

Peer’s style is very uneven and varies between the raw and the sophisticated. It is possible that the account has been written over a long period of time during which the writer himself has evolved. However, that does not absolve the responsibility of the publisher’s editorial team to play their role, which is more than spell checking.

Peer completely omits the Kargil war and is similarly silent about the Indo-Pak peace yatra that started with the Lahore bus trip by Atal Behari Vajpayee and has gone through its yo-yo moments.

These are glaring blemishes to ignore, especially from a trained journalist.

Peer stumbles often in maintaining distance and some circumspection regarding his own emotions and concerns. For example, there is too much shock expressed when a youth who is dandily throwing grenades and sniping armymen gets killed in an encounter.

Surely, Peer did not expect such elements to be given a medal by the army?

I am sure the militant himself was mentally ready for “shahadat”, even though youth are prone to feel temporarily invincible in the early stages of any insurgency. The fact of the matter is in such armed insurgencies there are very few armed men surviving till the end game (say in PLO or IRA).

Peer also exhibits a casual disdain for the changes that are occurring in India in the last two decades and rubbishes them with the label of a discredited “India Shining”, an affliction of many a blinkered anti-establishment writer.

In fact there is every reason to believe that these changes are also occurring at least in Srinagar and Jammu if not in rural J&K, albeit in a small way, and that is affecting the attitude of a section of Kashmiri youth (mostly born post-Gawakadal) who want to move on.

The fact that despite the hysteria of the Amarnath agitation in Jammu and Srinagar, the prime movers of the agitation on both sides viz BJP and PDP did not win either Srinagar or Jammu seats in the general election says something. There are long queues for recruitment into new BPOs opening up in Jammu and Srinagar.

Then again, the recent prolonged strikes in Srinagar post-Shopian and a suicidal destruction of the livelihood of hundreds of thousands of Kashmiris engaged in the tourist trade, tells us not to get carried away too much and that the old is still very much alive.

On the whole, Basharat Peer’s Curfewed Night is a welcome addition to contemporary history, written with passion and pathos.
It is surprising that we have so few of these in India (at least in the English language). Why don’t we have more such attempts to tell the story of Manipur, Nagaland, Narmada valley, the jungles of Orissa/ Chhattisgarh/ Jharkhand, Dharavi, Emergency, Amritsar ‘84, Delhi ‘84, Mumbai ‘93 or Gujarat 2002 in print or in film?

Why don’t we have our Norma Rae, Erin Brokovich or My heart lies buried at Wounded Knee? An Amu (Delhi 1984) or a Parzania (Gujarat 2002) are not enough.

Hopefully, more writers will follow Peer’s lead.