Grid Systems Enhance Processing

thePoint
January 6, 2005

Grid technology uses divide-and-conquer tactics to distribute computationally intensive tasks among any number of separate computers for parallel processing, reports Bank Systems & Technology. It allows unused CPU capacity—including, in some cases, the downtime between a user's keystrokes—to solve large computational problems.

While this technique has long been used to satisfy the insatiable computational needs at Wall Street's trading firms and investment banks, grid computing is taking hold in other areas of financial services, including insurance, corporate banking, and even retail finance.

Like the Internet, grid computing started in academia and the national defense industry and has moved into the commercial marketplace, along with its proponents. Phil Cushmaro used to work on grid computing applications for the defense industry, designing device drivers and operating systems for Concurrent Computer Corporation (Duluth, Georgia). Now, he's CIO of Credit Suisse First Boston (CSFB), the corporate and investment banking arm of the Credit Suisse Group (Zurich, $689 billion in assets).

Cushmaro uses grid computing as an antidote to IT complexity, and the approach could benefit other segments of the financial services industry. "You wouldn't allow your employees to remain idle 70% of the time,” he says. “And you shouldn't stand for it with your lazy computers, either.”

Drawing upon the work of the Globus Alliance, a technology consortium that publishes industry standards for grid computing, IT administrators can mix and match applications and processors.

"Now, you can basically take anybody's computer and anybody's software, and as long as it obeys this standard, it can participate in the grid," says Jason Bloomberg, analyst at Zap-Think, a Waltham, Massachusetts technology consultancy.

That's in theory, at least. True interoperability will take time. "Once you agree on a standard, you have to implement it, and once you implement it, you have to do interoperability testing on different products using that standard," says Lawrence Ryan, director of the financial services industry practice at Hewlett-Packard in Palo Alto, California. "There are standards today, yes, but these standards are also evolving very rapidly. They're not there yet."

That's why companies including HP, Fujitsu Siemens Computers, Intel, NEC, Network Appliance, Oracle, Sun Microsystems, and others have formed the Enterprise Grid Alliance of San Ramon, California, which will promote standards for enterprise grid applications.

Grids reduce complexity by de-emphasizing the importance of any given processor. A malfunctioning node in a grid doesn't bring any single process grinding to a halt. In fact, even individual transactions can make it through an outage unharmed.

Productive Screen Savers

In the commercial banking side, Wachovia Bank of Charlotte, North Carolina ($401 billion in assets) currently has a 300-node grid deployed for its Corporate and Investment Bank (CIB), and the technology is already paying real dividends. For starters, the bank has been able to reduce the overnight processing run for one of its critical applications to under an hour. Also, the grid provides significant quality-of-service improvements, to the point where 99% of its processing requests fall under a certain response threshold, up from about 85% before the grid.

The CIB grid is expected to grow fourfold over the next 12 months, and part of the growth will come from an interesting source: the idle computers of its employees. "After 10 minutes, if nobody has touched a keyboard, we say that [machine is] now available, and it's going to start calling for work," says Jim Kittridge, vice president of information technology for Wachovia.

It's called "dynamic provisioning," in which individual processing nodes register with Wachovia's DataSynapse grid middleware, which then matches all available CPUs with the currently requested tasks. But when someone returns and touches the keyboard, a user's machine will revert control to its in-person user. "We don't ever want to affect what the people are actually doing, whether they be traders or analysts across the bank," says Kittridge.

Wachovia's architecture standards board has approved an enterprise-wide grid strategy, according to Kittridge. "That does not mean we are going to have a single grid across the enterprise," he notes. "What it means is that we're all going to support a common architecture across the bank for where and when we deploy grid."

Grid Is Good

What's great about grid is its architectural flexibility. For instance, Sun Microsystems describes a core-processing strategy that uses nothing more than a networked collection of branch servers. "Say you've got 1,000 branches," says Dave Moore, global manager for banking at Sun Microsystems in Palo Alto. "You could conceivably have at least 1,000 [Java Branch Controller] servers sitting in those 1,000 branches, and each one of those has at least two processors."

Do the math, according to Moore, and you'll discover a grid powerful enough to perform the overnight core banking chores, thus allowing a bank to dustball its back-office mainframe. "Theoretically, we can take those very small branch processors, integrate them into a grid environment, and voila, the core system runs at night when the branches are closed," he says.

Also promising from an IT architecture standpoint is the Linux operating system. "It's the cheapest way to run a grid," claims Curt Burmeister, director of Linux and distributed computing solutions for Algorithmics in Toronto, a software provider for investment banks. Linux on Intel chips are "typically two to three times faster than proprietary Unix chips," he says. "It could be one-tenth the cost and twice as fast."

"That's huge," adds Burmeister. "We're talking millions of dollars in hardware budgets that are being saved."

But grid isn't a panacea for banks, cautions ZapThink's Bloomberg, as it's not suited for network-intensive tasks. "A lot of the things a bank does on a day-to-day basis are essentially transactions," he says. "Transactions, as a rule, are not particularly suitable for grid computing."

This story was written by staff at The Point for Credit Union Research and Advice and first published online at www.thepoint.cuna.org. Reprinted with permission.