The Tire-Kickers

These IT managers are giving on-demand a close look, but they aren't fully sold. Here's a look at their hopes for this new computing model, and their key concerns.

On-demand computing is a concept some IT managers have held up to the light and turned around a few times. But while the computing model has clearly captured their interest, they're not 100% sold on it. Like others in the IT field, they're taking a slow and cautious approach. In fact, industry experts say that while some users have already deployed applications that take advantage of utility computing for a specific function or business requirement, for most users, the road to on-demand will be an iterative process that will take years, if not up to a decade, to complete.

"I think we'll continue to see pilot projects in the short term over the next several years, with more enthusiasm and deployments building as more of the technologies necessary for utility computing come to market," says Jamie Gruener, an analyst at The Yankee Group in Boston. These developments include integrated data center automation and virtualization tools, easier-to-understand pricing models and improved capabilities to establish and maintain service levels across multiple technology layers within the data center. "It won't happen overnight," Gruener says, "but a number of larger customers have already begun the journey, with midsize customers starting to pilot programs this year and going forward."

Gordon Haff, a senior analyst at Illuminata Inc. in Nashua, N.H., agrees that on-demand will be a gradual migration. "We will reach a point where applications won't have to sit on this particular server with this much memory," he says. "Instead, there will be a pool of applications, and you won't have to worry about which server is actually running it."

Computerworld talked with three CIOs who are examining the on-demand computing model. Here's what they had to say about what they hope it can deliver, what they're concerned about and what it will take to convince them.

Harry Roberts
Senior vice president and CIO, Boscov's Department Store LLC

• Computing environment: In addition to a z900 enterprise server, Boscov's data center houses an RS/6000 running customer resource management, business intelligence and data mining applications. It also has a server farm made up of IBM xSeries PCs. All of these systems are connected to two 3.2TB Shark storage-area networks (SAN) from IBM and support about 700 users. Outside the data center, there are 2,500 desktop PCs (mostly IBM), 3,500 point-of-sale devices and a number of Unix servers running in-store applications.

"We're trying to position ourselves to be more open-source-oriented so that we can move our software to whatever architecture we have spare capacity on," says Roberts. "We're not quite there yet, but we hope to be in a couple of years."

• Hoped-for benefit: The ability to better align computing costs with the monthly revenue stream.

• Biggest concern: A lack of third-party software.

Like most retailers, Reading, Pa.-based Boscov's goes into peak computing mode two months of the year, during the November-to-December holiday shopping season. And like most retailers, the company pays year-round for the hardware and software to support that peak period.

Roberts is intrigued by the idea of on-demand computing, where the company would pay only for the computing resources it uses. "Our software is priced based on the capacity of the hardware we're using it on. So we're paying not just for hardware but for the software, based on our peak capacity," he says.

An on-demand model would not only theoretically reduce Boscov's IT costs, but it would also better align computing costs with each month's revenue.

But while Roberts sees great potential and has spoken frequently with IBM about its on-demand computing architecture, he still sees areas that need to be worked out before on-demand computing is feasible. For instance, there's the question of independent software vendors. "We haven't seen our third-party software vendors step up to the plate and have a complementary program to IBM's," Roberts says.

In addition, a move into IBM's on-demand architecture would require Boscov's to replace its IBM z900 server, since IBM's on-demand plan is focused around its z990. "To justify the necessary investment of retooling our shop, there would have to be some tangible percentage reduction in cost, and I don't know how on-demand is going to achieve that," Roberts says.

• What it will take to convince him: Roberts says he wants proof that on-demand will result in a lower total cost of ownership. "To date, there aren't enough details to make me comfortable that it will achieve my goals," he says.

Roberts also wants to see the third-party software vendors come to the table with licensing models that match on-demand architectures. "Everyone has to be in agreement," he says.

Jim Hull
Vice president of engineering services, MasterCard International Inc.

• Computing environment: MasterCard has three IBM z900s; several HP/Tandem servers for its debit applications; 700-plus Sun Microsystems Inc. Unix servers for its financial and settlement applications, Web server programs, Oracle Corp. finance programs and project management system; 1,000-plus Windows NT servers; and 1,500-plus IBM x205 series PCs. The Purchase, N.Y.-based company has tried to standardize on IBM and Sun servers. For storage, MasterCard has almost all EMC Corp. systems, with some from Hitachi Data Systems Corp.

• Hoped-for benefit: The ability to quickly add CPUs during the holiday shopping season.

• Biggest concern: Varying maturity levels from different vendors, and no consistent template.

Hull is also intrigued by the possibilities of on-demand computing but believes it's quite immature. As a result, he's cautiously implementing on-demand-related technology only in pockets of the enterprise where the architecture can accommodate it and where the business truly needs it.

Although the major on-demand vendors offer technology at varying levels of maturity, there's no on-demand template that's consistent from vendor to vendor, Hull points out. This makes users susceptible to vendor lock-in. "Let's say you chose Vendor X, and it turns out it was the wrong choice," he says. "What if on-demand took a different direction, but you've already bought into what they were selling?"

Another challenge is sorting through vaporware, like in grid computing. "Vendors say, 'We have it,'" Hull says, "but then you ask, 'What leading financial industry player is using it? For what applications? How does it scale? Does it work on technology that's three years old?' It turns out that conceptually it's great, but there's not a lot of people doing it yet."

MasterCard is using on-demand technology such as SANs, which allow servers to share pools of storage, as well as IBM's Capacity Upgrade on Demand on the zSeries 900. With CUoD, MasterCard can quickly add CPUs through a firmware download. This makes the holiday shopping season easier to handle; previously, upgrades had to be planned well before peak processing time.

The trouble with CUoD is that so far, IBM doesn't let companies turn processor power off, as Hewlett-Packard Co. does. In addition, Hull says, "since you may have CA or EMC or Sterling software running on the mainframe, as soon as you add the additional IBM engine, you have to pay for those other products as well."

MasterCard has also brought Sun's Solaris 10 operating system, which is expected to enable logical partitioning, into its test environment. According to Sun, Solaris 10 will enable users to easily move applications around on existing pools of servers, adding and subtracting processing power where it's needed.

"Companies have whole pockets of servers running at 10% to 15% utility, but if you can stack multiple applications onto one server, you can get better return for your dollar," Hull says.

• What it will take to convince him: Like Roberts, Hull says that he'd need to have a better understanding of how on-demand would be priced and how it would effect other software. "Particularly on the mainframe, where you have IBM, CA, Sterling, BMC and a whole bunch of other software—while it may make sense financially for the IBM piece, you might drive up the cost significantly on the others," he says.

There are two other areas Hull would like addressed: better industry standards for vendors to adhere to and a long-term commitment from the vendors that they'll continue to support the on-demand models they currently promulgate. In particular, he doesn't want to start down a road with one vendor, only to end up having the vendor change direction in three or four years.

Ultimately, Hull would like the entire infrastructure to recognize and respond automatically when something changes. For instance, if CPU use increases, he'd like the storage system to automatically respond, and vice versa, without human interaction.

"In a perfect world, I'd like to have a system that turns on additional CPUs in peak season, self-provisions and makes the service-level agreement, and when that was all done, deinstall the CPUs," he says. "The only way to do that is through open systems and open standards and people all driving toward the same concept of on-demand."

Mike Prince
CIO, Burlington Coat Factory Warehouse Corp.

• Computing environment: By fall, Burlington Coat Factory's environment will consist of thousands of Linux servers, many of which are IBM xSeries; some Windows desktop clients; Dell Inc. in-store servers; and a few Sun Sparc-based servers that it doesn't plan to retire. Its storage system is from Hitachi.

• Hoped-for benefit: A way to switch computing resources back and forth from general ledger operations at month's end to sales transactions that spike during the holiday shopping season.

• Biggest concern: Lack of automation to detect, monitor and activate a server.

Thanks to some old Sequent computers it had to retire, Burlington Coat Factory is pretty far along the on-demand computing path. "I don't think we'd have been as aggressive about on-demand, but we were faced with the need to make a change," says Prince. The Burlington, N.J.-based company had already moved its point-of-sale and in-store systems, as well as many functions in its distribution centers, to Linux, so it was logical to consider that operating system to replace its proprietary Unix servers.

Burlington Coat Factory is also a heavy Oracle user, employing not only its 9i database but also its ERP software and tool sets to build internal applications. With Oracle moving toward grid computing—which enables companies to dynamically move computing resources to the applications that need them—adopting that model made sense for the company.

On-demand was particularly attractive to the national retailer because of its dramatic dips and spikes in computing power needs. For instance, general ledger operations peak only at month's end. Sales transactions need tremendous amounts of horsepower, but only during the holiday shopping season.

Currently, Burlington Coat Factory has no way to switch computing power from general ledger to sales processing and back. As a result, it has traditionally bought much more computing resources than it actually needs. With on-demand computing, however, users have virtual resources in the form of clustered servers and can physically provision or deprovision those resources based on which application needs them at the time. "Instead of all those processors allocated to sales processing when you don't need them, you can allocate them to something else," Prince says.

The result: much lower overhead, and one-third to one-half of the capacity the company would need without on-demand. And because Linux runs on smaller IBM xSeries servers rather than on big-iron proprietary Unix, there are also hardware savings. "The price/performance is an order of magnitude better than the systems we're replacing," Prince says.

Burlington Coat Factory hopes to have its on-demand infrastructure in place by fall, including a leap to Oracle's 10g database. Other pieces of its virtualization puzzle include Cloverleaf Communications Inc.'s Intelligent Storage Networking System, which enables virtual storage allocation; Topspin Communications Inc.'s VFrame server virtualization software, which programs servers with the policies they need to determine when and how to cross-connect the processor, storage and I/O components to use shared resources; PolyServe Inc.'s Matrix Server, which enables servers to share storage as a single unit; and F5 Networks Inc.'s load-balancing software.

What's missing, according to Prince, is the ability to automatically detect, monitor and activate a server from a pool of spare resources. However, he's looking into technology from Vio Inc. to do that.

1 2 Page 1
Page 1 of 2
7 inconvenient truths about the hybrid work trend
Shop Tech Products at Amazon