Scientific Computing Associates

Financial Analytics: Lehman Brothers

Dr. Thanos Mitsolides, who today is Vice President of Fixed Income Derivative Products at Lehman Brothers, was charged by the company about ten years ago with finding ways to build cutting-edge applications involving lengthy and complex computations for risk analysis, P&L analysis, and portfolio exposure, with the additional complication that the applications be able to run effectively on conventional computers.

Dr. Mitsolides brought in Scientific's original Linda®, the pioneering development tool for distributed and parallel computing applications (recently superseded in Scientific's toolkit by the new TCP Linda^TM). "I wanted to parallelize several major applications and was looking for an efficient way to do it. My investigations showed Linda to be much better than many other supercomputing solutions," he says. "Linda was cost effective and easy to use. Since Linda is not based on proprietary languages or technology, it also provided vendor independence and allowed us to take advantage of the best price/performance ratios in hardware. I used Linda to write a toolkit and utilities that could be used by any of my programmers. Linda takes much of the programming complexity away, letting developers focus on the problem and the application."

About four years later, Dr. Mitsolides added Paradise®, the other tool in the LindaSpaces Computational Toolkit, to the Lehman Brothers IT infrastructure. Dr. Mitsolides found Paradise to be just as easy to use as Linda. Programmers at Lehman Brothers needed only about one day of training to take advantage of Paradise's advantages, such as the ability to assign tasks to available processors dynamically, balance processing loads, and restart or reassign tasks that have failed without affecting the overall application.

An example of the type of application that Lehman Brothers found could be enhanced by Paradise is the analytical work performed on 20 dual-processor Unix servers in New York. Paradise enables these servers to function as a computational pool - accessible by more than two hundred traders and analysts worldwide - for deal pricing and risk analysis of fixed income derivative products, many of which involve complex financial structures such as American-style swaptions and index amortized swaps. Large amounts of compute power are needed to perform efficient analysis.

With Paradise, computational tasks can be prioritized and assigned &qwuot;behind the scenes" in ways that are transparent to users. If tasks require more processing power than are available in the pool, Paradise can provide access to additional resources on an as-needed basis. In one such instance, a problem requiring complex analysis of extraordinarily high volumes of data exceeded the compute power capacity provided by Lehman Brothers' New York computational pool. The system administrator was quickly able to assign six additional servers from London's fixed income derivative operation to the New York pool, enabling completion of the analysis by the required deadline.