Until recently, grid computing was the domain of high-performance computing sites that calculate petabytes of data for particle physics problems or modeled engineering simulations. Now, grid computing is going mainstream, opening up new possibilities for carriers looking to support their growing application portfolios while keeping a lid on information technology budgets.By distributing processing workloads across networks of already existing computers, companies can attain higher performance processing power for relatively little additional investment.
Although grid technology is still a fairly new technology that's being tested by many companies, early reports from the field indicate it's viable and capable of delivering significant benefits, especially in deferred hardware costs and improved processing capabilities.
For example, J.P. Morgan calculates that it saved about $1 million in 2003-and will save even more this year-by implementing a grid that saved the bank money in hardware, software development and licensing, and IT maintenance costs.
Grid computing primarily is viewed as a strategy to cut costs in terms of hardware and software acquisition costs-but it also enables faster modeling for sophisticated actuarial analyses. In insurance and financial services companies, for example, analytical modeling is likely to be among the first sets of applications to run on grids.
Experts note that the best feature about grid computing is that it can run across any type of networked computers-regardless if whether the machines are servers or desktops, Intel-based boxes or mainframes, or running Windows, Linux, Unix or other operating systems.
Such configurations of diverse machines can be built into what Dan Kusnetzky, vice president at Framing-ham, Mass.-based technology consultant IDC, calls "virtual supercomputing pools," which can be assembled into scalable solutions of tens, hundreds, or even thousands of CPUs to meet the demands of processor-hungry applications.
A new IDC study shows that the grid computing market is on the verge of major expansion, expected to exceed $12 billion in revenue by 2007.
The technology is clearly breaking out of the high-performance computing space, Kusnetzky adds.
Taken to heart
One carrier, Genworth Financial Inc., Richmond, Va., has taken this approach to heart and has pressed its network of desktop workstations into service as a grid to achieve close to supercomputing power at PC prices.
Genworth was spun off in early 2004 from GE Financial Assurance Holdings Inc., the holding company for GE's principal U.S. life and mortgage insurance operations. Genworth offers life insurance, long-term care insurance, payment protection insurance, retirement income and investment products, and mortgage insurance. With more than 15 million accounts in 20 countries, processing demands are huge, and the company needed additional computing horsepower for its actuarial number-crunching operations.
Prior to its grid implementation, Genworth supported actuarial analysis and modeling with a cluster of 10 servers, according to Scott McKay, senior vice president and CIO for Genworth Financial.
Genworth has a relatively small IT staff, and the carrier's systems were difficult to maintain and were due for replacement. In addition, while clustering dynamically increased processing power in line with workload increases, adding additional nodes would have required an expensive license upgrade in addition to new hardware.
To meet increasing actuarial processing needs, Genworth linked 100 of its desktop PCs into a grid architecture, employing GridServer software from DataSynapse, New York. The PCs vary in age from brand new to up to three years old. The typical PC has a 1.5-GHz processor with 256 MB of RAM.
"Our business is very data-intensive and very risk-focused," McKay explains. "Grid computing helps us with the analytics we need to manage the business from a risk perspective, as well as to help with our product management.
"We needed an ability to run our analytics in far shorter cycle times, and therefore allow us to run a significantly higher number of scenarios to increase our effectiveness in managing risk. But we needed to do this in a way that's cost effective and asset effective," he adds.
The cost to deploy processing across a grid was nominal compared with the expenses associated with building out the clustered configuration, or adding new server boxes, McKay notes. "We have deployed desktops that are already assets in use within the company," he says. "We're increasing the effective utilization of those assets."
The DataSynapse software is installed as a small executable program on each workstation, and is managed from a central server that dynamically distributes the processes to underutilized processors. McKay reports that while some large jobs will employ up to 90 computers, most use up to half that amount.
IDC's Kusnetzky is not surprised that companies such as Genworth are turning to PC networks for grid solutions. "Most organizations have desktop systems that are largely under-utilized," he notes.
Many of Genworth's PCs were underutilized, and therefore represented a considerable reserve of idle processing cycles, McKay says.
"We looked at the jobs of our people working in actuarial and finance, and measured how much time they interacted directly with their computers, versus how much time they were on the phone or in meetings. We found that the effective utilization of their desktops was fairly low," he explains. Other prime candidates included machines sitting in conference or training rooms, he adds.
Although McKay met some initial resistance to the idea of replacing the actuarial department's server clusters with a PC grid, the positive results changed users' perceptions. "When we first told the actuaries that we were retiring the servers and going to a grid, they were not exactly thrilled with the idea," he says.
Actuaries were pleased with the system's vastly improved processing times. "The actuaries had asked for a three-times improvement in the cycle time of running some of their models," McKay says.
"With the servers, the actuarial models were running over-night, and some took as long as 24 hours. So in some cases, it's going from 24 hours down to 30 to 60 minutes to run the same job."
The grid also has improved Genworth's risk management capabilities.
"Our actuaries are able to do more sophisticated analysis, more stochastic scenarios, and other types of modeling," he explains. "They can do better analysis, because they now have the computing power that they didn't have before at their fingertips."
End-users' workstations tapped for the grid experience virtually no performance degradation, either, McKay notes. Typically, because grids can dynamically route work between underused processors, there is little or no performance impact on individual workstations, and Genworth is no exception.
"You can actually set a tolerance with the software," he explains. "For example, you can set it such that if the user even moves his or her mouse, that it takes that unit of work that it was processing and sends the work to another engine. It's very unobtrusive."
IDC's Kusnetzky agrees that this dynamic rerouting is a prime advantage of grids. "Applications are unobtrusive to desktop users-they are not even aware of the emulation environment running on their machines," he says.
Security is also addressed by grid software, he adds. Software can be configured so users do not have access to the applications' underlying code and data on their desktops nor access to the desktop users' data.
With the success in its first phase of grid, Genworth is planning to deploy additional actuarial applications across this processing pool. "We're looking at additional actuarial applications," says McKay. "We've got two or three actuarial investment applications that we're looking at."
McKay's team also is looking at running some of its core business applications across a grid as well. In the process, Genworth may begin to expand out to server-based grid implementations.
"Some of the applications that we're looking at as part of this assessment are ones that would not run on a desktop grid, so we're looking at a server-based approach also," he explains. "We're now doing an assessment of available CPU cycles on our servers."
While Genworth has about 5,000 workstations at various locations, McKay has no immediate plans to extend the grid beyond the company's headquarters. Even some local staff may have some limitations, he adds.
"Some people are heavy users of their PCs, and others are in service centers processing transactions all day. Some of those desktops would not be candidates for this. We also won't use machines used for hard-core analytics."
Nevertheless, McKay finds no downside to a grid deployment. "Every CIO should be looking at grid computing," he advises. "It's a double win from a cost standpoint. You get it up front by retiring servers and leveraging the hardware that you already have in place. If you're doing a better job of managing risk as a life insurance business, you're also winning on the back end, because the policies you're selling are better underwritten, and therefore more profitable.
"The amount of additional power we're getting from the grid is tremendous," he adds. "With the grid, we're seeing anywhere from 10 times or more improvement in cycle time. Just by retiring dedicated hardware and leveraging assets that we already have."
Joe McKendrick is a technology writer based in Doylestown, Pa.
Register or login for access to this item and much more
All Digital Insurance content is archived after seven days.
Community members receive:
- All recent and archived articles
- Conference offers and updates
- A full menu of enewsletter options
- Web seminars, white papers, ebooks
Already have an account? Log In
Don't have an account? Register for Free Unlimited Access