dcsimg
 

Blade Servers: Evolving for the Enterprise

Monday Apr 21st 2003 by Carl Weinschenk

When first introduced, blade servers were lightweight servers on a card that were fast, cheap and out of control. Today's blade servers are still fast, but they're designed for more complex tasks and can be better managed.

Blade servers have only been around for a few years, but they already have an interesting history. While few people take issue with the rationale behind then, the technology had the misfortunate of arriving just as the business sector into which they were borne was decimated.

The reaction wasn't to meekly depart the scene, another good idea falling victim to tough times. Instead, blades have been reinvigorated by a strong move into the enterprise sector. They have been repositioned to perform a wider variety of tasks. The result has been a vote of confidence by some of the biggest vendors in the communications industry.

Blade servers can loosely be defined as servers on a card. The strength is in the elements they share, which include the backplane, fans and management. This saves valuable floor space, reduces the amount of cabling and provides great management flexibility. It also enables great scaling: Blades can be scaled simply by adding cards to the existing chassis.

"We see a blade as a single board computer," says Ashley Eikenberry, the group manager for blade product marketing for Sun Microsystems. "The whole benefit of blades is that you can start to have network elements that you can put side by side like books in a bookshelf."

During the past 18 months, the industry's chief concern -- and thus the key driver of blades -- has been the driving down of total cost of ownership (TCO). "Clearly in the IT market today the first, second and third issue to the CIO and CTO is how to do more and spend a whole lot less," says Susan Davis, the vice president of marketing and management for Egenera. "All the dynamics are toward reducing TCO, from capital expenditures to ongoing management."

A Strong Segment
The move into the enterprise data center has occurred as the original market for blades -- telecommunications companies -- has been decimated by the economy. The strength of the blade concept is shown that despite the slump in telecommunications, the big vendors have moved strongly in this sector. The segment appears strong, with a good mix of blade-specific and general computing companies on board. Blade-only players are RLX Technologies and Egenera. The big players are Sun Microsystems, IBM, Dell and Hewlett-Packard.

Virtually all the blade action is in the enterprise. "Blades are the next data center," says Paul Barker, the vice president of marketing for RLX. "Just as we saw the made up over time of computer power, storage and networking, all those elements are moving onto the blades."

The use of blades in the enterprise differs from their use in telecommunications. Eighteen months ago, says Rob Sauerwalt, the global blade center marketing manager for IBM, the key was density. Data-center designers and CIOs were preoccupied with the amount of power used by each blade. The higher the power, the more heat that must be dispelled and the fewer blades that can be packed together. The result, he says, was low-wattage (and, thus, low performing) blades. This was an appropriate approach for implementations in which the tasks were simpler. "This is fine if you are running apps that are not very demanding," he says. "It played into the ISP boom."

Today, however, the goal is to sell blades into medium and large enterprises with much more complex computing requirements. "These are more traditional businesses," Sauerwalt says. "The dynamic has shifted to standard class performance solution and a higher level of functionality. They are losing the 10-watt processors and looking for more traditional platform levels."

A key driver of TCO is the ability to use the devices flexibly and, for that reason, get by with fewer servers, each of which operates at a higher level of efficiency. Eikenberry says that estimates that standalone servers operate at 15 to 20 percent of capacity. Blade servers, when run under an efficient management umbrella, can run at 80 of capacity, she says.

The reason that efficient is so much higher in a blade environment is that the integrated management enables them to be flexibly delegated to specific tasks on an as-needed basis. A side benefit is that management is far easier, according to Eikenberry, because the compute power appears in a single profile to the end users. Management of dozens or even hundreds of blades can be performed with a simple interface. This management approach -- called "virtualization" -- is extremely attractive to enterprise managers, especially in an environment where trained IT personnel are at a premium. "Having a solution for system management is amazingly important," Sauerwalt says.

Getting on the Grid  
Virtualization is in essence the execution of futuristic approaches to computing in which the physical location of the server matters less than their ability to be linked logically. Blades, then, are the technological approach underlying such futuristic approaches as grid, pervasive and cluster computing. Though these approaches differ somewhat, they share some traits, such as a high degree of flexibility. "Blades are the enablers of grid and [other] dynamic infrastructures," says Paul Miller, the director of blade marketing for HP.

The integration of blades is rising to the top of the priority list, says John Humphreys, a senior analyst with IDC. He says that there are three key issues driving blades: Reducing power consumption, increasing the density and integration. Integration -- the ability to easily managed perhaps hundreds of servers -- has recently grown to be considered the most important virtue of blade designs by enterprise IT managers, he says. "Power and density fell further down on the list," he said.

Humphreys says that there two levels of management. Each vendor generally has a proprietary level. Some, he says, buttress these with standards-based management such as simple network management protocol (SNMP)-based tools. The two most common are IBM's Tivoli or HP's OpenView.

Insiders suggest that the near-term future of blades will be characterized by increased standardization, higher capacity connections between blades and from blades to servers and other initiatives aimed at increasing the ability to virtualize and otherwise improve blade flexibility. And, while the business is still in a downturn, blade vendors remain optimistic. "Blades are leading the convergence of computing, networking and storage into a single infrastructure environment with a consistent management system," Miller says.

 

Vendor/Product Introduced OSes Management Density Storage
Sun Fire B100s 2/2003 Solaris Proprietary/Standard 16 services in 3u 30 GB onboard disk
Egenera 10/2001 Red Hat, Linux, Windows (2Q) Proprietary 24 blades in 42u SAN, NAS support
HP ProLiant BL40p 1/2003 Windows, Red Hat, SuSE Standard 2 per 6u, 12 per rack 584 GB internal, SAN
RLX ServerBlade 2800i 2/2003 Windows, Linux Proprietary 70 blades, 140 processors per 42u rack

60 GB drive, 8 GB DDR, NAS support

IBM  eServer BladeCenter 11/2002 Linux, Windows Proprietary 14 per 7u rack 80 GB IDE, 147 GB SCSI, SAN, NAS support
Home
Mobile Site | Full Site