JP Morgan says disposable computing is key to IT

JP Morgan Chase has solved the problem of how to provide computationally intensive derivatives pricing and risk modelling capabilities across products and throughout the firm, without vastly expanding its IT budget with what Steve Neiman, head of high-performance computing at the New York-based bank, calls a “disposable computing infrastructure”.

Speaking at the Waters 2002 financial and risk management technology congress this morning in New York, Neiman said it rapidly became apparent that the cost of buying specialised hardware for each derivatives product line was prohibitively high, especially since the firm’s appetite for computing capacity steadily grows, and systems designed for one product often do not work well for others. “Risk management systems are always asked to do more and more – more volume, more complexity, more everything,” he said. In particular, modelling exotic derivatives – which Neiman said comprise roughly 8% of the firm’s derivatives business – is very computationally intensive.

To achieve a scaleable and cost-effective solution, the firm has exploited the plummeting cost of generic computer hardware to build what it calls a “compute backbone” comprised of scores of identical computers that can simply be thrown out and replaced when they fail – and with hundreds of them working in concert, some fail every day, Neiman says.

The key to the system’s success is a fault-tolerant layer of software that sits between the hardware itself and the object-oriented software tools used to build and run the risk management systems, Neiman said. This layer offsets the lack of fault-tolerance in the hardware by distributing the computational tasks among the computers and re-routing around problem units, in an approach borrowed from old-fashioned batch processing computer systems. This keeps costs down. “Building fault-tolerant hardware is extremely expensive; fault-tolerant software isn’t,” Neiman noted.

One delegate asked why the firm used a dedicated computer farm, rather than using so-called peer-to-peer technology to exploit underused desktop PCs. Neiman responded that JP Morgan Chase’s risk modelling needs would rapidly outstrip that capacity.