Grid Computing: The Grid Is Going Mainstream Fast

The systems have come far from their roots, and are now maturing. Ohio Savings, Wachovia and CSFB are getting their feet wet, as the GSEs use it to clean up regulatory trouble.

By Michael Dumiak

Grid computing may still sound fringe to buttoned-down comptrollers looking over technology budgets at small Midwestern thrifts. They've got a point-grids came into being out of academic research labs and NASA's need for physicists.

But when a mortgage player like Cleveland's Ohio Savings starts talking about the grid, maybe it's not just for the space age any more. The firm's enterprise information manager, Tony Miller, says Ohio Savings is using a grid network now at the database level, running an $800,000 Oracle 10g system. Even so, he and project manager Ed Kizys figure soon to move the vital retail loan and mortgage-origination systems onto the grid, putting it right at the center of the firm's reason for being. After all, if the likes of Goldman Sachs are investing in grid-computing software companies-then grids must be something financial companies can use.

A wide variety of financial companies are starting to get on board. Grid computing is basically about distribution and power-sharing, collecting servers into nodes, spreading computations among multiple machines. "It really comes down to speed, doesn't it?" says Steve Manning, vp of computer networks at mortgage giant Freddie Mac. And sure, grids are seen more often these days among Wall Street shops and options players. A big trading desk breaks up calculations into small pieces and distribute them along a grid of computer nodes. Then the individual answers are reassembled into a final result. The nodes collect the power from various computers to gain efficiency and use otherwise-idle capacity, gaining more power and speed. The idea conjures warehouse-sized server farms and UNIVAC research labs-IBM just opened a Deep Computing On-Demand grid for rent to financial companies, for example-but it doesn't have to be that way.

"All the servers sit in one rack," Miller says of his setup in Ohio. "That includes tests and production." Ohio Savings is a $12 billion operation; the system offers more than a terabyte of power. Miller says Ohio Savings' grid is made up of a three-node cluster. The firm is using its Oracle grid, running on Red Hat Linux on HP DL380 servers, to carry its call center for retail customers and on-line banking system, part of its internal portal and voice-response unit telephone system. It ties back in to a storage environment provided by EMC. "None of these are heavily loaded systems where we have a lot of concurrent usage," Miller says. "This was basically done to prove out the 10g environment, to make sure it worked."

As opposed to the more down-to-earth Ohio players, DataSynapse software is running on grid systems at massive institutions from Wachovia to Credit Suisse First Boston. It's running where one might expect a need for huge numbers of calculations fast-for risk management, for the options and fixed-income desks of worldwide trading operations. It's about the speed. Wachovia's fixed-income group, for instance, used to run 4,000 risk simulations a day for options, and those numbers would have to be crunched overnight. The processing power in grids allow 20,000 to 100,000 simulations daily. Chief executive Peter Lee says speed is compelling, but it's not everything. "Another part is availability," he says. "Uptime. When you move to a fabric of hundreds of nodes, or even dozens of nodes, one going down doesn't end your system."

It was a good enough proposition to interest the elite investment bank Goldman Sachs, long known as a technology innovator. After spending a year investigating grid technologies, Goldman decided to use DataSynapse software in its global risk management operation. It also decided to sink money into Lee's company. But it's not just risk and derivatives driving grid, which is seeing a rise in popularity along with Linux adoption-making the likes of LinuxWorld program director Dave Rosenberg declare that "grid has gone mainstream." It means more efficient use of servers, distributed systems allowing for greater security, financial players are adopting service-oriented architectures that dovetail nicely into grids.

But for Freddie Mac? Necessity was the mother of invention. Two years ago Freddie, along with its sister mortgage market-maker Fannie Mae, ran into serious accounting issues-Freddie wound up having to restate earnings from 2000 to 2002, finding that executives underreported income of $5.3 billion to make its numbers smoother, resulting in a flurry of heads rolling and $125 million in civil penalties. Fannie overstated its income by $11 billion, forcing high-flying chief executive Franklin Raines to step down. "It was to solve problems internally," Freddie vp Manning says about the firm's introduction to grid, a little wryly. "There was a lot of number crunching."

About the same time the firm ran into trouble, Manning had been hearing from software vendors about grid technology and, once a small test had been done, was amazed at the results. Freddie built its system in-house. "In reviewing the financials I can't tell you how many models we had," he says. "We had one that was running for a month that we ran in about a day when we moved over to grid. We just wouldn't have been able to get things done." Manning's talking about default capital modeling for the mortgage giant, by which the tracks its portfolio when taking on assets or running them off; these inputs constantly vary the default likelihood of the portfolio, changing the amount of capital Freddie must keep on hand to meet regulatory safety requirements against a crash. It's a massive amount of information to track.

Freddie runs its grid on two data centers in the Washington, D.C., area with between 500 and 1,000 nodes, depending on need, each between one and four CPUs. They run on Intel boxes of various shapes and sizes, mostly HP and IBM.

For Manning, the goal of grid is speed translated into faster answers for more questions-from pricing and evaluation, to risk and to whether a trader will buy loans. And he sees a lot of room to roam. "Right now the grid is in a data center," he says. "We really want to start using the desktop assets." He notes unused desktops can be patched into a grid to "lend" their idle power. "I've got 9,000 PCs sitting on people's desks at any one time. Our next goal is to make those part of the grid."

Still Andrea Klein, the financial services industry strategist at software giant Oracle, expects grid technology to reach further into everyday financial uses. The most immediate-and perhaps much further-reaching than the fixed-income desks of Wall Street-is replacing the aging software in cash machines. Just now those millions of machines are underpinned by even bigger machines-IBM mainframes and the like. "That marketplace is ready to be churned. It's high-cost," she says. Oracle technology marketing vp Robert Shimp says grids are less complicated to install and hardware vendors are simplifying the wiring and processing that grids need; software layers are also being integrated into single stacks.

Klein is bullish on grids running Linux, which she argues runs better on those systems than Microsoft Windows-based software. With money machines becoming more sophisticated and processing needs increasing, however, Klein may have her finger on a point where even the thrift down the road takes a look at running grid-based software.

"You're going to see a lot more features and functions because they're going to be able to deliver more to the machine," Klein says. "The features and functions you'll be able to do at the ATM itself will be far more varied. It takes a lot of power."

http://www.banktechnews.com/article.html?id=20050701DAJNGF4J