In-Depth

Consultancy Puts IBM’s TCO Claims to the Test

Do Big Blue’s mainframe TCO assertions add up?

• By Stephen Swoyer
• January 2, 2007

IBM has pulled off a clever marketing strategy over the last 18 months. The company has managed to recast its System z mainframe line as a cost-effective alternative to conventional Unix, Linux, and Windows servers.

IBM kicked its mainframe TCO counter-offensive into high gear in 2006, sponsoring research by consultancy The Robert Frances Group (entitled "Mainframe Role TCO") which cited mainframe TCO advantages in the form of reduced power and cooling requirements. "IT executives and their staff should consider the mainframe as a 'Tier 1' option for hosting new applications and acting as a central hub for security, server pool management, and consolidated workloads/data," the researchers wrote.

Robert Frances isn’t the only industry watcher that’s suddenly gotten hip to the TCO benefits of mainframe ownership.

Consider a recent study by Wayne Kernochan, an analyst with consultancy Illuminata Inc., which found that mainframe licensing costs—the historical thorn in the side of mainframe shops everywhere—have now become cost-competitive with other platforms.

"A key reason for this is IBM’s focus on driving these costs down. For example, IBM’s database-administration tools have had the effect of driving the prices of competitors such as CA and Compuware down significantly," Kernochan writes, adding that IBM’s licensing costs typically decrease per unit of workload as workloads themselves increase. IBM doesn’t stop there, of course: System z officials claim that the mainframe enjoys a TCO advantage of between 5 and 60 percent over typical Unix, Linux, and Windows alternatives.

These are heady claims, to be sure—and Illuminata recently put them to the test.

For starters, Kernochan acknowledges, IBM correctly breaks down TCO into hardware, software licensing, people (e.g., administration, implementation, and upgrading), and environmental (e.g., electricity and air conditioning) costs. This, Kernochan notes, is fairly standard stuff.

On the other hand, he points out, IBM’s mainframe-versus-distributed-workload comparisons are anything but standard. This isn’t necessarily a bad thing, however. "[I]nstead of considering single-application TCO scenarios, as is often done, IBM assesses cases where tens of applications are running, either on a single mainframe or on multiple distributed Linux/Unix or Wintel platforms," he comments, noting that this is a typical usage model in medium-sized to large enterprises.

For the record, IBM has published TCO results that range from only marginally advantageous (e.g., System z versus System p) to as much as 60 percent better. "System z wins on the basis of administrative costs," Kernochan points out, noting that mainframe administrative costs are simply much lower than are those for competitive platforms.

"[I]ts software costs are about the same, and its hardware costs much more. The number of administrators required to run the operating systems, hardware, and networks of the distributed systems increases nearly linearly as the workload scales, while a single mainframe’s administrative costs barely increase at all."

The upshot, Kernochan argues, is that "mainframe people costs are often a fraction of those costs required for distributed systems."

Illuminata does quibble with IBM’s approach in some respects, however. For example, Kernochan suggests, Big Blue’s highly consolidated mainframe workload model almost certainly understates the importance of administrative costs—for application, infrastructure, and (especially) database management. For example, the more database instances you consolidate on a single system, the more DBAs you’ll have to have on hand to effectually manage them.

"In a world where just four Oracle instances frequently require a dedicated administrator, database administration costs can reach $10 million in a three-year period for an application with 50 Oracle copies," he points out.

The reverse of this argument is that these costs are universal. "[T]hese costs apply to any platform, and therefore do not affect the relative TCOs of the mainframe, Unix boxes, and PC servers," Kernochan indicates. "[I]n spite of [IBM’s] glossing over this particular variable, it remains a valid point that mainframes are often cheaper—even considerably cheaper—to administer in a large, mixed-workload environment."

On the hardware side, Kernochan allows, mainframes are still significantly more expensive than their RISC/Unix and volume server brethren. This picture, too, is much improved, at least relative to five years ago. The mainframes of today, after all, are considerably cheaper than their Big Iron forebears.

"[W]hile still quite pricey by volume-server standards, the price of mainframe memory and other components have been significantly chopped as well, with continuing aggressive price cuts planned," he observes. "IBM’s focus on cooling technology also means that the mainframe should typically require less electricity and air conditioning than many 1U servers running the same workload."

The salient takeaway, Kernochan concludes, is that IBM’s TCO claims have merit. "Add it all together and mainframe TCO can look quite attractive. Administrative costs are often lower—even much lower—and … hardware, software, [and] physical plant costs are [not] … the drag they once were."