Re-Engineering J.P. Morgan
Why High-Performance Computing Can Be A Bear
February 7, 2012
That happened to be where J.P. Morgan had a major development center. Windows were broken throughout the company’s 75-story tower, downtown. Nine hundred technology and operations staff members were displaced. So were 800 other staffers covering 600 lines of investment banking business. “It was a nightmare,’’ said Cherasia. With staff already displaced, “Lehman was an absolute stop dead for the derivatives position migration.’’
But ultimately about 2 million derivatives positions got winnowed to 1 million, through the Strategic Re-Engineering Project.
All told, the three-year project is expected to generate $300 million a year in savings to the bank and $1 billion of business benefits.
Plus, it’s allowed Cherasia and his team to start thinking that maybe they, and not Goldman Sachs, are becoming the best technology and operations team in capital markets.
Among the core developments of its re-engineering event are its derivatives and securities trading platform, known as Athena.
The system relies on a classic form of computer science programming known as an “acyclic dependency graph” that interconnects risk calculations, securities prices, durations of prices, yields, cash flows and reference data.
What the graph allows Athena to do is determine at every moment just what “leaf in the tree” has to be recalculated. This makes for what Cherasia calls “lazy recalcs.”
“Things only need to be recalculated when you need to recalculate them,’’ he said. Processing cycles are saved.
The technology team also has built its platform out of Python, a general-purpose, high-level programming language whose design philosophy emphasizes code readability. The reference implementation of Python, known as CPython, is free and open source software.
This allows J.P. Morgan to recruit talent already familiar with its basic programming language. That “allows our quants to come in and build their models directly on the platform, very quickly,’’ Cherasia said, “which allows us to leverage those products broadly across the platform.”
The code that gets produced, in turn, is all connected to an object-oriented database that is replicated in real time across every one of its major trading centers, in what Cherasia calls “ring fashion.”
This lets quants and technologists, he said, to work directly with traders to solve problems – and deliver updates every day.
In addition to the re-engineering project, J.P. Morgan is pursuing high-performance computing, to reduce cost and speed of bring services to market, while improving performance.
This includes ‘’virtual server’’ farms and the use of advanced chip technologies, including field programmable gate arrays (FPGAs) and graphics processing units (GPUs). This improves the speed and efficiency of running quantitative models.
J.P. Morgan has accelerated much of the work that ends up on its “virtual servers” through the use of the graphical processing units. These are the kinds of computer chips that first got widely used in video games, where speed and complexity, as well as constantly changing colors, backgrounds and interactions are paramount.
Such chips modify the style of the processor to meet the style of the code, Cherasia notes.