7. Scale-Up: Big, centralized servers, often mainframes. In general, mainframes trade-off easy manageability and power for potentially difficult migration or upgrade paths. As such, they are considered an outdated idea in some circles.
Confusion emerges when the major vendors, which, with the exception of Unisys, tout both scale-up and scale-out offerings, use the same terminology to praise scale-out and scale-up in separate places. If the vendors haven’t clearly chosen between strategies, how should the users? Consider, for example, the wording of this Unisys recent press release, titled, “Unisys’ ES7000 Meets the Scale-up Requirements of the PeopleSoft 8 Internet Architecture.” It notes, “companies are now faced with a proliferation of small and midrange systems that are incompatible, unable to share information, and difficult or impossible to manage consistently.” The document goes on to say, “Nearly all major server suppliers, notably HP, IBM, and Unisys, are talking about server consolidation.” It then defines server consolidation as “the process of replacing those small and midrange distributed servers (“scale-out”) with highly scalable or “scale-up” servers.”
Forrester’s Gillett sees scale-up as viable, albeit shrinking, proposition, “The data center will continue to have scale-up, but I’m betting that it’s a shrinking percentage of the total capacity of the data center because of the compelling economics of Intel and the two inexpensive operating systems, Windows and Linux,” he concludes.
8. Grid Computing: A collection of scale-out computers stitched together to, in essence, form one super-computer. Despite much hype, grid computing has made few inroads in the enterprise and remains the domain of science and university applications. When looking at new and emerging technology like grid computing, Gartner’s oft-referenced “Hype Cycle” is a useful model to determine the current practical value and deployment window. Its “Hype Cycle for Servers, 2003” (published in May 2003) cites “Growing confusion over definitions, benefits, maturity, and applicability” for grids in the enterprise. Despite hype to the contrary, it argues, pragmatic commercial grid applications are as much as five to 10 years away. One reason for confusion might be that the term grid is “sometimes misused to denote the related technologies of distributed and utility computing.” But alas, the report concludes: “Little is known about what commercial grid applications might be.”
9. Utility computing: A hypothetical environment in which a number of scale-out computers and IT resources can be redeployed on the fly to meet changing business needs. Phrases like “on-demand computing” and “Adaptive Enterprise” are synonymous with utility computing: All are feel-good terms created to persuade buyers that redeployable technology is the solution for all that ails. On one side, most technologies have utility, are somewhat adaptive, or can be redeployed “on demand,” as needed, thus rendering the terms a bit useless. On the other, lack of true interoperability between technologies and operating systems can lend these terms the hype-only frustration of grid computing. The natural offshoot of these “plug and play” type ideas is that the IT department becomes commoditized, possibly encouraging enterprises to outsource their entire IT organization or take it offshore.
10. Total cost of ownership (TCO): The true cost of owning a server, including IT management, power, accessories, repairs, and the time spent taking the Fujitsu service guys out to lunch. In theory, TCO is a truly useful concept to consider when making a server purchase. In practice, however, even the most obviously expensive vendor can claim that its product reduces TCO based on a given factor. Consider a recent Web-based ad for Microsoft Windows 2000, which asserts “Windows server offers a savings of 11 percent to 22 percent over Linux in 4 out of 5 workload scenarios.” The ad links to a Microsoft “Facts” site, where a Microsoft-sponsored study asserts “Lower Windows Staffing Costs Provide a TCO Advantage over Linux.”
NIKA Consulting’s Stephen Adams sees this differently, and backs it up with a real-life example. He describes a mail deployment where a company “needed to support 50-plus offices and 1,600 people with e-mail.” He notes, “Using Exchange required 50 servers and four full-time staff ” Using Linux, Adams claims, “it could have been done with fewer than five servers, and supported by no more than two people.”
Microsoft’s fact bent is obviously pro-Microsoft; Adams’ is pro-Linux, thus proving our point that TCO can be used in basically any argument.
But we’ll trust the imparitial guy in the field.