A calculated risk

• CDO risk simulation involves hundreds of thousands of calculations.
• Larger banks have their own grid systems, allowing them to outperform smaller competitors.
• Commerzbank's New York office is running a CDOSheet grid across its PC network.
• Risk analysis of a CDO portfolio, which would normally take days to run on a single PC, was completed in less than an hour in a proof of concept on Sun Grid.
• Simplified pricing - cusomers pay for exactly what they use.

By Gillian Law, 25 August 2005

GRID computing is about to shift the balance of power in the investment banking world.

Until now, large banks have had access to risk analysis data that just wasn't available to smaller banks and investment companies - thereby making those smaller companies dependent on their larger cousins for their information. Grid computing, and specifically pay-as-you-go access, is about to change all that.

The catalyst for this change is Sun Microsystem's Sun Grid, and its first licensee, CDO2, a UK software development company based in London.

CDO2 focuses on running simulations for popular credit derivative funds called CDOs, or Collateralised Debt Obligations. Director Gary Kendall set the company up eighteen months ago, when he spotted a glaring gap in the market.

"CDOs are structured in a very complex way that lets people spread risk around. Where a normal share would relate to one company, a CDO includes hundreds of companies in each security and the investor takes on only a slice of the risk. And then there are CDOs-squared, where lots of CDOs themselves are bundled together. In order to look into all those companies, and price the CDO, you have to run many hundreds of thousands of numerical calculations to get an estimation of the value," Kendall says.

Large banks manage this calculation by having their own grid systems in place, churning through the figures overnight to calculate the risk for each CDO. Smaller companies, however, struggle. Due to the complexity of the calculations, many don't have systems in place at all. And those that do, often only get an approximate view of what is going on because of their limited resources. This means that they often have to rely on, and trust, the information given to them by larger banks.

"It's a very small niche area, but I saw a market there for using grid computing to run those calculations," says Kendall. "And the customers don't care that it's grid computing - they're just amazed and surprised when I show them what the software can do, and how fast."

Some banks are already running CDO2's CDOSheet software in-house, such as Commerzbank's New York branch. "The Commerzbank example is different, in that we have created a grid using a network of existing desktop PCs, and the calculations are run on those," Kendall says. "But in future it will be much easier for clients' work to be run on Sun's Sun Grid. Because we've written all the code in Java, we have been able to take that exact same software and run it on the Sun Grid without any difficulties."

"You have to consider that some investors in CDOs, such as small hedge funds, only have a network administrator for technology support", Kendall added. "For these customers, running a grid in-house to get valuations is not a viable option."

CDO2 and Sun recently completed a proof of concept (POC) showing the benefits of running CDOSheet on Sun Grid. They ran it on a pre-production test facility - a128 node (256 CPU) Sun Grid using the Solaris 10 operating system, based in a data centre in Linlithgow, Scotland.

Risk analysis of a CDO portfolio, which would normally take days to run on a single PC, was completed in less than an hour, the companies say.

Live customer transactions will of course be run on the ‘real' Sun Grid, which is spread across four centres around the world, Kendall says. Sun launched its pay-as-you-go grid in February, but CDO2 is its first working client.

Working with Sun Grid makes life simpler for Kendall, because he can leave the running of the grid to Sun, and also charge users by the CPU cycle.

"That has completely changed the market, the ability to move to pay as you go," he says.

"It opens the product up to a whole raft of customer who could not have considered using grid computing in the past," he says.

Sun charges US$1 per CPU per hour plus $1 per gigabyte per month for storage and CDO2 charges a premium on top of that, he says. "And there's a monthly maintenance charge, but that's a flat fee, regardless of how many users they have. So it really means people can commit to using it, knowing that any time they need to run more complex simulations they can afford to do so," he says.

CDOSheet uses the dynamic networking capabilities of Jini technology, allowing resources to be added and subtracted as needed while the software is running - so that customers use, and pay for, the computing power they need and no more.

CDO2 and Sun are setting up a customer test centre in Guillemont Park, Camberley, where potential customers can come and test out the software on a small 32 CPU rack grid.

"It's computing power on demand, when companies need it. I spotted the need for this, and there's already strong interest," Kendall says. "Also, some imminent changes to international accounting standards mean that banks can't rely on their competition for data - so this is the ideal opportunity to offer them an alternative they can afford," he says.