High-Performance Computing Revs Up at BMO

Every time somebody signs into their Facebook account, they access a server farm of more than 10,000 computers linked together in a High-Performance Computing (HPC) environment. At Google, a grid of 450,000 machines processes each search.

These virtual supercomputers, with their massive and economical computing power, are being adopted by financial institutions – including BMO Capital Markets, the investment and corporate banking arm of BMO Financial Group – to evaluate risk in complicated and interrelated financial markets.

In areas such as Credit Derivatives and Risk Management, risk evaluation is now so complicated that calculations can take days to solve. For example, it takes 170 hours – more than seven days – to calculate the financial exposure to BMO Capital Markets’ more than 1,000 derivative counterparties. Finding a way to speed up this analysis could give BMO’s traders a key advantage.

The software development team of Byron Bianco and George Cawkwell worked with hardware expert Sorin Manta to develop BMO’s first internal grid computing environment. Rather than follow the Google example, which is loosely coupled and geographically scattered, they created a centralized, high-density cluster environment. BMO’s grid computer uses high-performance fibre optics running between the computers, and so can be viewed as a single entity. “This gives us extremely high performance,” says Manta. “Counterparty risk reports that would have taken a week to generate can now be done overnight.”

Adding to the impact HPC is having on the organization is the fact that this computing power can be made available to any part of the firm that needs it. It can be dynamically allocated to any task being performed intra-day or overnight. It also offloads the Excel processing load from individual desktops, placing it into the computer grid. This speeds up both the processing and the local system, while ensuring proper controls around the spreadsheet.

Manta and the team worked with Microsoft and HP to assemble the best-performing and most cost-effective solution available. As the first financial institution in Canada to deploy this type of architecture, BMO enjoys direct access to Microsoft’s engineering teams, and is contributing directly to the future direction of this technology.

“Having a partnership with the platform manufacturers gives us a clear view of future improvements to the cluster architecture,” says Manta. “In 5-10 years, we expect to see teraflops capacity on a single chip, which means that mainframe-scale processing power will be available on inexpensive platforms.”

There has been enthusiastic response to the HPC capabilities from all over BMO Capital Markets. The Financial Products Group, for example, is developing an in-house derivatives pricing library, the Mathematical Finance Library. Currently run by traders on their local PCs, the library will soon be using HPC for the faster run times required to support intra-day trading decisions for the most complex derivative portfolios within the bank.

Jeff Poulsen, global head of Structured Product Trading, welcomes the growing role of technology in his group’s work: “The gap between BMO’s traders, model builders, and platform teams must continue to close if we expect to be competitive and profitable in the capital markets environment. The HPC team has shown that they can provide quick, robust and flexible models, and we are very excited about what that allows us to accomplish.”

Technorati