In-Depth

IBM's Latest Blades Show Slim Is In

All grown up, blade servers are seen as a complement to mainframe and large Unix systems. Now they're slimmer than ever—just one of recent trends in this hardware.

• By Stephen Swoyer
• 01/13/2004

The upshot is that server blades have come a long way since they were forced into enterprise environments—as low-power, edge-of-network devices—and are increasingly being tapped in multi-node clustered configurations as replacements for large Intel-based servers and as complements for large mainframe or Unix systems.

Through 2001, companies generally used blade servers—typically low-voltage Pentium III-based systems of 1 GHz or less—to support edge-of-network applications. In this respect, IBM’s new BladeCenter HS40 illustrates just how far the industry’s come, packing four 2.8 GHz Xeon MP chips into a server blade less than two inches high. All told, up to seven four-processor HS40 blades can slide into a 7U IBM eServer BladeCenter chassis. Big Blue’s not alone: among major vendors, Hewlett-Packard Co. beat IBM to market with its BL40p four-way blade server by about a year, although—not surprisingly—Big Blue claims that its four-way is the slimmest blade yet.

“About 18 months ago, blade servers were a different thing, usually a single low-powered processor, but things have changed a lot since then,” says Jay Bretzmann, product marketing manager for IBM’s eServer xSeries line. “Now you’re seeing hundreds of blades clustered and handling a lot of workloads that you used to see only on big SMP servers.”

Typically, clusters of server blades are used in high-performance computing (HPC), technical computing, or as part of server consolidation efforts. In the last case, many companies are tapping scalable server blades to move off of highly-distributed, aging Intel-based systems, industry watchers say.

“These are mostly physical consolidations where you just take a whole bunch of technology that’s beyond its warranty or support and you’re looking out for the latest and greatest phenomenon, which are blades,” says Bretzmann. “Not only can you fit a whole lot more servers in a whole lot less space, but you gain a lot of other advantages, such as reduced cabling, rapid deployment, and better manageability.”

Still another trend, industry watchers say, is that customers are transitioning away from large Intel-based SMP servers and toward more modular blade architectures. “The customers in large SMP environments using eight- or 16-way [SMP servers] have done it in a way where they want to provide the absolute best protection, so they’d buy two eight ways, or two 16 ways,” says Anthony Dina, manager of business development for blades in HP’s Industry Standard Servers group. “But with a lot of the advances we’ve seen in clustered file systems and being able to break those jobs into smaller nodes, we’re seeing greater use of the IT assets. So if you look at it from the perspective that it’s cheaper to do it, and it’s a dramatic improvement in the cost efficiencies, it helps to explain the popularity [of server blades].”

In late 2002, IBM shipped its first blade servers, dual-processor systems that also featured fibre channel connectivity for storage area network (SAN) and other applications. Since then, other vendors, including HP, have followed suit. As a result, says Steve Gillaspie, blade systems manager within HP’s ProLiant group, blades are increasingly being tapped to support multi-tiered applications. “There’s been a lot of SAN growth in the enterprise, so blades are being deployed for more than just edge-of-network applications,” he points out. “We’re seeing installations of SAP, PeopleSoft, Oracle—a lot of the traditional multi-tiered applications that you’d see on a traditional large SMP server.”

Viewed in the context of overall server shipments, blades still account for a small percentage of the total. Through the end of Q3 2003, for example, Market research firm International Data Corp. (IDC) reported that 120,000 blades have been sold. At the same time, IDC analysts pointed out, 50,000 blades had been sold in the third-quarter alone—a strong indication of growth in a server market that is still rousing itself after nearly three years of hibernation.

For his part, HP’s Gillaspie claims that some of his company’s customers have deployed “thousands” of clustered server blades, and representatives from both HP and IBM say that clusters of 32 or more blades are common. Clustering in the blade space can take a variety of different forms, including load-balancing—primarily used for Web; e-mail or other IP-based applications—message passing for HPC and other applications; as well as availability, in the form of failover clustering solutions and clustered file systems for SANs.

HP and IBM have also introduced a variety of administrative enhancements designed to make the job of managing dozens or hundreds of server blades easier. HP ships Rapid Deployment Packs for both Linux and Windows that allow administrators to rapidly deploy Linux and Windows systems on blades. According to HP’s Dina, administrators can use the Rapid Deployment Packs to quickly repurpose existing blades to handle seasonal changes in demand, or other spikes in volume. “It includes both a graphic user interface and a command line interface and the combination of that with the scripting ability allows customers to get servers quickly up and running,” he comments. “If the line of business needs more assets or more equipment to run peak workloads, they can start migrating some of the mail and messaging and other servers to those new roles, then use the [Rapid] Deployment Pack to return those to traditional tasks when these seasonal changes are over.”

IBM, for its part, markets IBM Director, a tool for its xSeries systems that lets administrators view and track the hardware configurations of remote systems in detail or monitor the usage and performance of critical components. Director includes configuration wizards and has a software distribution option that can help with the configuration and deployment of server blades.

Some industry watchers outline a scenario in which server blades will continue to cut into the market share of traditional large SMP servers, but all agree that there’s a place for both. For starters, a recent report from research firm Gartner Inc. found that although server blades generate comparatively less heat than rack mounted servers, they consume more power and have greater heat dissipation requirements. These factors, considered in tandem with their greater cost—few, if any so-called “white box” vendors are shipping dense server blades—could limit their acceptance.

Adds HP’s Gillaspie: “Large SMP systems will still be very important for what we would term a scale-up architecture for very large databases, so they would be used in conjunction with our very large servers or blades, but they would be very focused on those very high-end mission critical applications.”

IBM Corp. has championed the use of blade servers alongside large mainframe and Unix systems. “A two-tier approach, what we’re calling a ‘Scale Out and Scale Up approach’ can help customers better manage their technology assets,” Pete McCaffrey, IBM’s program director for zSeries marketing, told us last year. In the “Scale Out” approach, McCaffrey explains, customers can tap blade servers that allow them to add computational capacity by slapping in additional blades. “Scale Out” approaches, on the other hand, include mainframe and high-end Unix systems that can add additional capacity on demand.

“There’s sort of the counterintuitive notion of mainframe+1, the mainframe sort of representing the ultimate in scale up, whereas blade centers are sort of the ultimate in sort of a scale-out environment,” McCaffrey said. “We’re seeing customers getting the two centers of gravity and in some cases tying them together to match the application to the right technology.”