Utility Computing Still to Come

By Maria Trombly | Securities Industry News | March 29, 2004

March 29, 2004 - Despite benefits, firms still hesitant to outsource certain functions

Many people like the idea of utility computing. Companies shouldn't have to manage their computer hardware, the same way that they generally don't need to run their own coal-burning power plants or wind farms.

Now plenty of vendors are ready and waiting to take over entire IT departments. And that's useful for some non-core IT processes. But there are certain applications that a Wall Street firm might not want to hand over to an outsourcer. It would be nice to pull in computing power when it's needed, but keep the applications and data in-house.
Unfortunately, that's where the metaphor of computing power and energy grids breaks down.

Electricity does, in fact, enter your company. It does its business, then it goes away. You can pull in as much as your breakers will allow, or use none at all, if you remember to turn the lights out when you leave the building.

Computing power, however, cannot move over a wire. The computer is where it is, and all the calculations, computations and processing take place right there where it lives.

You can ship data and instructions over, and get results back, but the actual computing power does not, in fact, flow into your company like electricity does. And it's unlikely to do so anytime soon.

Instead, the phrase "utility computing" has come to mean one of two things-business process outsourcing and in-house utilities.

Business process outsourcing takes a complete corporate function--hardware, software, data, even people--and moves it under a vendor's control. Sometimes, but not always, the process is moved to the vendor's facility, as well. This could be a problem for Wall Street firms.

"If you're a J.P. Morgan or a Merrill Lynch, I'm not sure you want to share that much of your infrastructure," said Joe Hogan, VP of marketing for HP managed services at Hewlett-Packard. "Unlike electricity, inside of that infrastructure, you have a lot of proprietary data, like sensitive information about your clients, so you need to have some degree of privacy."

As a result, Hogan said, he sees very limited applications for utilities outside the enterprise for securities firms, other than for such applications as e-mail and messaging.

"So what we're seeing now is not the ultimate [computing power] grid but the grid within the enterprise itself," he said.

J.P. Morgan Chase is the pioneer in this area, combining seven separate financial risk management systems to share computing power using a grid computing system. The project was built using grid tools from Platform Computing. The bank estimates that it has reduced computing costs by 20 percent due to lower costs for hardware, reduced development costs and easier system management.

In general, utility computing appears whenever an institution moves beyond the level of isolated, single-department grids.

"When you have a number of departmental grids growing into a multi-departmental grid or enterprise grid, as soon as you start to do that, you have to start thinking about the pricing model," sad Bob Boettcher, VP of financial services at Platform Computing.

The hardware and its management still represents a fixed cost for the IT department that runs the grid, but computing becomes a variable cost to the departments that use it.

"This is really the first stage of utility computing," said Boettcher. In this scenario, all data and applications still live within the walls of the enterprise, but computing power can be more efficiently allocated.

Grids tend to operate at higher capacity than computers in general, and variable pricing can drive that even higher. For example, low overnight rates can inspire some departments to move their less time-sensitive applications to those hours.

"One derivatives unit at a large investment bank is able to access its grid cheaply, because they don't mind doing it off-peak," Boettcher said. "So they're able to do many, many more computations. They're able to relax all sorts of assumptions in their models, because they have almost free computing power."

Eventually, Boettcher said, companies will get so comfortable with internal utilities that they will be more ready to adopt external utilities, though that is still a few years off.

Other companies implementing in-house utility grids are Goldman Sachs and Credit Suisse First Boston, according to Susan Davis, VP of product marketing and management at Egenera.

She added that she's seen a lot of progress over the last few months. "Wall Street, as usual, is leading the way," she said. "Especially for things that are very compute-intensive, like trading systems and analytics work, with the goal of optimizing the use of their resources. It's a way to share computing resources and dramatically drive up utilization. The result of that is that you buy less because you're taking advantage of more of the dollars that you spend."

If You Build It, Will They Come?
Not every in-house utility is the same as far as departmental users are concerned. If the utility is used to add an additional server or storage space when it's needed, then departments don't have to rewrite their applications to take advantage of the utility. But if the utility works by distributing applications to dozens, or hundreds, of networked computers over a grid, then the department might have to do some work to make the application grid-ready.

IT departments that build grids need to be ready to sell their grids to users, and help users get their applications ready for the grids, said Eliot Listman, CEO at Powerllel Corp.. Otherwise, those grids might wind up sitting idle, he said.

"You need to help those business units adapt applications for a distributed environment," he said. "How do they break up the application to be sent to the grid? How do those applications get load balanced? There are lots of technical issues that need to be overcome to get those applications to run."

This is another area in which the power grid is completely unlike a utility computing grid-the power coming from your generator is basically identical to the stuff coming from electric outlets. But applications do have to be rewritten for them to continue to work over a distributed network.

Fortunately, some vendors of analytical applications are already adapting the software so that it can work on grids if needed. For example, in cooperation with Powerllel, SciComp has developed a parallel-processing version of its SciMC, a derivative pricing application.

Overflow Utilities
One intermediary step towards utilities-both internal and external-is to use utilities just for overflow capacity.

"That's essentially what we're doing," said John Hasson, relationship director for central services at Abbey National Plc. He has two grids set up, one a small, well-defined set of dedicated computers used for risk modeling. Another, a more flexible grid composed of underutilized employee PCs, is used to handle overflow processing tasks for a pricing application.

According to Hasson, his firm typically buys much more computing capacity that it needs, in order to make sure that there's enough capacity for peak periods. "That leaves a lot of troughs," he said.

Setting up a utility grid allows him to reclaim some of that wasted computing power. "I would like to see the same thing extended to the use of storage," he added. "We would typically use 60 percent of all the storage capacity that we buy. Leveraging our networked storage--that's the direction I'd like this to go. So then computing really gets to be a utility." The overflow could also, in theory, go to an outside vendor.

"We recently finished a test where the job was sent to IBM to finish," said Frank Cicio, chief marketing and strategy officer at DataSynapse. To do this, the offsite utility--in this case, IBM--had to be ready to take the application's overflow. In Cicio's words, it had to be "preconditioned."

"We knew that if we ran out of capacity, we would use IBM's data centers," he said.

Utility Pricing
One big unresolved issue with setting up external utilities is how to calculate utility pricing rates. With an in-house utility, a firm can pretty much decide on its own standard for charging departments and go with it. But for external utilities, a company has to be able to compare prices from different vendors.

"We are in discussions with some large Wall Street firms about hosting," said Don Russo, Oracle's VP for financial services.

But the hosted price is based on a percentage of the license fee, not on the number of computing cycle the customer would use, he added. Oracle doesn't yet have a utility pricing structure because of the difficulty of coming up with a pricing model, said George Demarest, senior director of database marketing at the company.

"You basically have to instrument all the applications, the storage, the servers and say, you use X numbers of units," he said. "This is no simple task. There are no industry standards, and in order to get to that utility pricing, you have to get industry agreement for how to measure that."

Oracle is also struggling with a pricing structure for in-house grid licenses of its products. "Grid computing will present different requirements for licensing and we are working on it but have not announced it," Russo said.