Blade Servers: Big Iron in a Small Form Factor
Blade server success founded on key mainframe concepts
It’s a now-familiar scenario: Two prominent researchers disagree about the specifics of a market—in this case, that of blade servers—and two tier-one vendors (i.e., Hewlett-Packard Co. and IBM Corp.) both fall over themselves trying to claim a share of the bragging rights.
In this case, the real winner is the blade server space, which continues to post impressive growth. There are a variety of reasons, analysts say, not the least of which is that blade servers—for all of their modularity, configurability, and inherent manageability—boast many of the same features that have made the mainframe relevant for 40 years.
Rack servers have been with us for a long time, but commodity blade servers, at least, are a post-millennial development. Think of blades as expansion cards—complete with one, two, or even four processors, memory, and hard disk storage --- that plug into a chassis. In this way, blades can be clustered in much greater densities (numbers of processors) than conventional rack-mounted servers.
HP and IBM have only been marketing blade systems for a couple of years, but are already locked in a struggle for market leadership. For the record, HP topped IBM in research firm Gartner Inc.’s latest quarterly survey. IBM, on the other hand, topped HP in research firm International Data Corp.’s latest quarterly survey. Because both vendors claim to lead the market in unit shipments and overall revenues; both—perhaps rightly—claim the blade server market crown. The same thing happened last year, when Gartner and IDC split over relational database market figures.
Blades are Hot, Hot, Hot
The latest research data, while sharply divided over blade server market hegemony, does coalesce on a few key points.
First, blade servers are hot. Gartner finds that blade servers are the fastest growing segment of the server market, for example, and HP last week announced that it had sold 100,000 blade servers in slightly more than two years. In fact, HP has sold approximately 50,000 blade servers in the last seven months. At this rate, the computing giant expects that it won’t take nearly as long to hit the 200,000 mark.
“We’ve seen quarter-over-quarter growth that’s very rapid here, so the projections to get to the next 200,000 will probably be much shorter,” confirms Steve Gillaspie, group marketing manager for blades with HP’s Enterprise Server Group (ESG).
What makes blade servers so popular? Sally Stevens, director of blade server platforms for HP, says that “Customers are looking to be more efficient in the way that they run their data centers.” The attraction of blades is that they’re managed by highly flexible tools—HP’s Client Consolidation Infrastructure (CCI), IBM’s Director—which enable a variety of very rich scenarios, most of them involving, to one extent or another, consolidations of existing applications or resources.
“There are a lot of choices customers have when they look at doing consolidation, and blades tend to be a natural form factor fit for this, not only from a server perspective, but also looking at storage consolidation,” she comments. “So they can put multiple applications on a blade and be able to consolidate. From that perspective, too, you’re not only saving floor space and power, but you also have fewer servers to manage, and in any event you have tools at your disposal [such as CCI] that let you manage them more effectively.”
It’s no coincidence that mainframes took their greatest beating during the Age of Plenty—the mid- to late-1990’s—when capacity was cheap and utilization (in most organizations, at least) little more than an afterthought. In the mainframe space, processing horsepower can be virtualized into very small slices of overall capacity, called logical partitions (LPARs), which host applications or jobs. This is important because few applications actually require the processing power of the big, discrete mainframe microprocessors. As a result, applications or jobs can be assigned a fraction of compute power, which means that system operators can husband the capacity of their systems and drive their utilization rates to near 100 percent.
For a time in the 1990s, this value proposition inexplicably ceased to be compelling. Enticed by the allure of cheap and plentiful hardware, organizations populated their environments with commodity servers, many of which were deployed to support tasks (such as file and print or e-mail services) that required only a fraction of their processing power.
Starting with the crunch of 2001, however, unused capacity—as realized in the dozens of standalone servers scattered around most organizations—became an albatross of sorts for IT. After foundering in the waters of excess, organizations were marooned with tens, hundreds, or even thousands of under-utilized and highly distributed servers, all of which required expensive upkeep in terms of human intervention, physical space, power, and cooling. Is it any wonder that the mainframe—which addresses all three issues in one (relatively compact) platform—has been on a roll of late?
The same goes for server blades, which are mainframe-like in many ways. Their sheer compact size lets customers group them in densities that—in spite of their commodity Intel or AMD processor innards—rival the capacity of large mainframe systems. (Ironically, IBM’s BladeCenter systems are powered by cooling technology that Big Blue originally developed for its zSeries mainframes.)
Like mainframes, blades are enabled as much by hardware as they are by software. Among tier-one vendors, HP and IBM have evolved their manageability tools to support common administrative tasks (such as operating system roll-outs, application installs, or rapid system rebuilds) as well as more sophisticated activities, such as the allocation or de-allocation of existing capacity, or the provisioning of new capacity to meet business needs.
Have daily fluctuations in your compute requirements? No problem, says HP’s Stevens. Just reassign capacity from one application to another. “Let’s say you’re using one [blade] as a Web server, but you don’t get much activity during certain hours, say, from 12:00 to 6:00 AM,” she explains. “You can assign that capacity to another [application] that needs it, like a payroll application, which is usually pretty busy during those hours.”
HP and IBM are also delivering technologies (HP by partnering with VMWare, IBM with its “Utilization Engine”) that purport to drive up blade utilization rates. The result, some analysts say, is that blades are compelling alternatives to standalone servers. “Among other things, they’re an alternative for carving up a single, larger system. Certainly at one level, and for some types of applications, putting in a rack of fairly lightweight blades is certainly an alternative to carving up a small number of big servers,” confirms Gordon Haff, a senior analyst with consultancy Illuminata.
Does this mean that blade servers are ready to replace Big Iron in enterprise environments? No one we spoke to (outside of Dell, HP, or other IBM competitors, at least) believes blades are or should be considered replacements for Big Iron. If anything, says Haff, it’s the conventional standalone server that’s the endangered species. “For a variety of reasons, I think there’ll continue to be a market for standalone servers, however. I’m not sure what the timeframe is, but if you look forward, in a few years, at any rate, you’re going to see blade-type designs really largely take over from the way rack-mount systems are built today, essentially for the cable management and for the greater density and for the ease of repair, and that kind of thing,” he predicts.
Stephen Swoyer is a Nashville, TN-based freelance journalist who writes about technology.