Blade servers: A thin proposition?

Blade servers have had a short and inauspicious history, primarily because they lack two things: standards, and a clearly articulated marketing message. The lack of standards in lamentable, and is unlikely to be successfully addressed any time soon. A clearly articulated marketing message may be less of an issue, but it is symptomatic of the fact that when blade servers first appeared, nobody really knew what they were for.

The attraction of blades is that each server is housed, usually with a hard disk, on a single card so that many -- in some cases up to 20 -- servers can share the same chassis, supply and connectivity.

Early marketing efforts pitched blade servers at ISPs as a replacement for the "pizza-box" market, but there were two problems with this: first, ISPs prefer to buy capacity as new business arrives, rather than buy in bulk which, initially at least, blade servers dictated; and second, ISPs were getting no new customers.

The 4321 rule

Against this background of cynicism, however, blade servers have found homes in an increasing number of businesses. IBM's xSeries general manager Susan Whitney puts the benefits in terms of the "4321 rule": blade servers are four times cheaper, consume one-third the power and half the space of a normal server set-up, which would produce the same performance.

Lastminute.com is one company that has cut its teeth on blade servers. As server load doubles year on year, the travel firm looked around for a resilient solution and, given the choice of scaling up to bigger boxes or scaling out with commodity parts, the company chose the latter route, according to head of technical architecture Paul Chudleigh.

INow most of Lastminute's customer-facing systems are hosted on 56 IBM dual Xeon blades in four Bladecenter chassis. Benefits to Lastminute have included being able to keep the systems running even when several blades are out of action, and ease of management.

It's deployments like this that have helped drag blade server sales to a $119m business in the second quarter of 2003. Although this is a 693 percent increase on the same quarter in 2002, the figure pales in comparison to sales of other types of server. Of the big vendors, HP, which was the first to launch product, had the largest market share at 31.9 percent, followed by IBM on 26.9 percent and Dell on 15.1 percent.

Fast blades in a slow market

Dell appears the most restrained about the potential of blade servers. Reza Rooholamini, a director in Dell's Enterprise Systems Group, says demand is not as strong as once perceived. She believes the real take-off in demand for blades is a year away. Illuminata analyst Gordon Haff sees Dell's position as a barometer of where the volume market is. "Blades have been a slow market to develop," he cautions.

Others are more optimistic. HP and IBM are currently locked in a race to squeeze the most processing power onto individual blades without having to resort to water cooling. In August, HP introduced dual-processor blades that use Intel's latest 3.06GHz Xeon DP processor and put a faster 2.8GHz Xeon MP chip in its four-processor blade. IBM is poised to release blades with 3.06GHz processors next month and plans to follow this with its first four-processor blades.

And while HP can augment its Intel blades with PA-RISC blades running HP-UX, IBM plans to ship blades based on its PowerPC 970 chip to early customers in the fourth quarter, with full production beginning in the first quarter of 2004 -- a delay from the original schedule, which envisioned volume shipments this year.

Sun's first blade system meanwhile, which lets 16 blades with the company's own UltraSparc IIe processor fit into a 5.25-inch-tall chassis, shipped earlier this year, but its second blade-server design, using AMD Athlon processors, is expected by the end of 2003. A dual-processor x86 blade is also on the cards at Sun, as is a dual-core UltraSparc blade -- the company has refrained from saying whether this will be based on the Sun's Gemini processor, a design using UltraSparc II processors intended for low-end servers.

When the x86 blades arrive, customers will be able to mix them with UltraSparc blades in the same chassis. It's a practice employed by one Sun blade beta customer, Canadian telecommunication company Telus, according to Craig Richardson, the telecom company's assistant vice president for hosting and managed applications.

Management is key

Squeezing more processing power into a smaller space is only part of the story. Sun is also adding "speciality blades" that plug into the same chassis, said Mangar Dange, Sun product line manager for network and security products. Sun is selling one speciality blade with a dedicated chip to balance loads among groups of blades, and plans another with a special chip for accelerating Web transactions encrypted with the SSL (Secure Sockets Layer) standard.

Sun's blade-management software is a first incarnation of its N1 technology to group equipment into vast, easily reconfigured pools of computing resources. The N1 software for blades makes it easier for administrators to quickly reconfigure the job assigned to one or several blades.

At Lastminute, IBM's management console allows one person to manage all 56 servers: the last thing the company wanted was to have to increase staff costs to manage the vastly increased number of servers.

"Increasingly software is the differentiator," says IBM's Whitney. "If I consolidate my software into a blade centre, I need software systems management tools to monitor and reprovision the servers. IBM has that. At somewhere like Dell, software is not a core competency."

There remain several challenges both for those selling blade servers and for those buying them. The first is cost. Although individual blades are relatively cheap, the cost of a chassis, plus management software and a few blades can be significantly more expensive than other solutions. IBM has made concessions to companies wanting to invest in blade architectures by introducing new pricing policies: customers can take delivery of a chassis with 14 blades, but only pay to have seven of those switched on. The others can be switched on later.

Standards, standards everywhere…

The second -- and less easily surmountable problem -- with blade servers is the lack of standards that would allow a blade from one manufacturer to work in a chassis produced by another. Nearly every vendor has issued calls for standards, but each appears to be a lone voice and all have so far failed to gain allies among the other vendors.

In a speech at OracleWorld recently, Michael Dell called for common blade architecture: "If we are successful, and I am reasonably optimistic we will be," he said, "I think you will see a high-volume market for these blades." Dell likens the current status of blades to the pre-Internet period in which computers were networked together using a variety of competing technologies, such as Ethernet and Token Ring. Only when Ethernet became ubiquitous, did the Internet, and computer networking in general, take off, he said. But he declined to say which businesses the company is talking to about blades or which components it is seeking to standardise.

Meanwhile, IBM has already formed an alliance with Intel to standardise the construction of blade servers, and HP, which in 2002 called for a standardisation effort around CompactPCI, was roundly snubbed by other server manufacturers.

It's a problem which, if fixed, could bring great benefits to those buying blade servers, if not those making them. "There is a lack of industry standards around blades at the moment which means that people who invest in the technology are locked into one vendor. But over the next four to five years we think there is going to be a great deal of evolution and blades will have some real potential in the right scenario," says Gartner analyst Andy Butler.

In the long run, says Gartner, we'll be seeing a lot of blade servers in datacentres of all size. In a recent research note, the analyst firm said that for the next year enterprises should look at blade servers as tactical solutions and require a short-term (two-to-three-year) return on investment on blade purchases. "Enterprises should anticipate more mainstream usage of blades by year-end 2005," it added, "with the assumption that blade server vendors will demonstrate progress on the steps outlined in this research. Finally, enterprises should expect blades to be ubiquitous technology in most data centres by year-end 2008 -- once again, assuming that blade server vendors continue to make progress on their road maps. And part of that means standardisation."