44 India's Best IT Implementations of the year 2005

Continued from page: 4

Friday, June 03, 2005

Most Technologically Challenging PRISM National Stock Exchange of India

Print

Comment

Email

Digg

Del.icio.us

Reddit

Twitter

Project Head: Satish Naralkar, MD & CEO, NSE.IT Ltd 
Location: Mumbai

PRISM or Parallel Risk Mgmt System is a fine example of using parallel computing to meet the business needs. Parallel computing is generally used in super computers for performing complex scientific calculations, but it's been used here for doing risk management in the stock market. The NSE (National Stock Exchange) deployed this project for its wholly owned subsidiary, the National Security Clearing Corporation or NSCCL. This subsidiary is responsible for the clearing and settlement of all trades executed on the NSE. When the futures and options market opened in India, there was a need to develop a risk-management system for it, which could monitor the risks associated with each investor. The computation of risks in this form of trading is quite complex, and if NSE's trading members violate any norms while doing so, they can be barred from trading any further. It's therefore not only important for trading members to know of the risks involved in derivatives trading, but they also need to know it in real time to avoid disablement. Moreover, any system that provides this information can't afford to drop even a single transaction, because that's equivalent of one trade being missed. If that single dropped trade turns out to be a risky one, imagine the consequences. That's where the PRISM or Parallel Risk Management system comes in, which provides this information to all trading members in real time during intra-day trading hours of NSE.

Building such a system is a technological challenge given the complexities involved in risk computations, which is why PRISM gets this prestigious award. The parallel computing architecture is normally used in super computers for running scientific applications. PRISM is a live example of using the same architecture for a business application. 

Typically, to create such a parallel computing system, it would have required mainframes or super computers. This would have considerably shot up the cost of building such a system. However, PRISM was developed using utility hardware, comprising a few 4-way Intel Xeon-based servers running RedHat Linux. The software for it was developed completely in-house using ANSI C, MPI or Message Passing Interface, Java, RMI and Perl. MPI is a standard communication interface used in super computers. Specifically the MPICH implementation of MPI was used, which is freely available. As the system's been developed in ANSI C, it can be deployed on any UNIX-based system, hence RedHat Linux was used. Java was used to build the front end of the system. 

Plus, the methodology used to calculate the initial margin requirement for each investor was SPAN or Standard Portfolio Analysis at Risk. SPAN identifies the overall risk in a particular portfolio for futures and options contracts for each trading member. It uses 16 standardized definitions of risk scenarios, which are defined based on the probable change of price over a day, or probable change in price volatility over a day. These help identify the largest loss a portfolio is likely to suffer during a day. 

How it works
The entire parallel architecture of PRISM is divided into four processes. These are called Received, GUI handler, Mother and Child. The Received process, as the name suggests receives live data from NSE's trading system. The GUI handler handles all GUI queries and user interactions. 

The mother does basic portfolio handling for traders, maintains the margin data and also monitors for all the trading violations. Plus, it does the crucial task of child process scheduling, ie finding a free child process and routing requests to it in real time. 
The child process is responsible for doing the margin computation. In case a child process crashes due to some reason, the mother will automatically assign it to another child process. Until a child process is assigned, the computation load is shared by the remaining child processes. 

The best part of the system is that its highly scalable. Any number of mother and child processes can be added to the system, based on the transaction load. Blade servers can be used for scaling the system up, since they can pack more computing power into lesser space. Currently the system can perform 200 computations per second, which is quite significant. 

Further still, the system has built-in software-based fault tolerance, instead of hardware-based fault tolerance to save costs. The fault tolerant system is identical to the live system and runs in parallel with it. In case the live system fails, control is automatically shifted to the fallback system, which then starts using the child processes. 

Both systems compute positions and monitor violations, but the disablement messages are only sent out by the live system. Such a system could not have been built with technical knowledge alone. It required a mix of software development skills in Java and C, as well as functional skills in derivatives and risk management. The real challenge, as claimed by NSE was in ensuring that the system worked in real time. 

The beneficiaries
The original beneficiary of PRISM is the trading community as it provides them online results to prevent their exposure limits from exceeding. This reduces the risks in this business, which could otherwise be very heavy. Returns offered by the system are instantaneous, as NSE handles 99.5% of all derivatives trade in India.

The system was tested at CDAC's supercomputing facility, and can be used in areas other than risk computation as well. It's actually brought the cost of super computing down, and can be used for other business applications like decision support, surveillance and trading engines. 

Overall, PRISM is technologically a very challenging project. It's a combination of ingenuity and complexity. Ingenuity is in developing such a system in house using utility hardware and freely available super computing standards. Complexity is in using the SPAN algorithms in real time for risk computation. 

Next Page : Maximum Business Impact GIS-based Risk Mgmt System ICICI Lombard

Page(s)   1  2  3  4  5  6  7  8  9  10  11