Two years after vendors began introducing slimmed down "blade" computers as the next stage in the evolution of the server, these denser, more power-efficient servers have been struggling to meet market expectations and prove that they have what it takes to emerge as an industry standard.
In fact, the last couple of years have been downright rocky for blades. The dot-com-fuelled Web serving market they were first aimed at has shrunk, and the IT market in general has grown much more conservative in its approach toward new technology.
Still, as far as analysts and blade vendors are concerned, the question is not whether these new servers will take off, but when.
Their confidence is understandable, because in principle, the blade concept makes a great deal of sense. The idea is to strip down the servers in a server farm to their bare essentials: removing the CD-ROM, video processor, and keyboard connection and having the power and cooling supplied by a separate enclosure, so that the blade servers can be slid in and out of specialized enclosures like books in a bookshelf.
In theory, you have servers that are less expensive to produce, take up less space, and require less power.
In reality, you have a brand new architecture with benefits that have been harder to prove in practice.
This is why blade sales have fallen short of analyst expectations. In 2001, IDC predicted that 67,000 of the slimmed-down servers would be sold the following year, creating a US$148 million market. The industry research firm was bullish about blade growth, forecasting sales of $4.5 billion by 2005.
By the time 2002 was over, however, vendors shipped 40 percent fewer blades than IDC had expected, and tallied just $100 million in sales, almost $50 million less than expected.
IDC has now nearly halved its 2005 sales prediction to $2.5 billion, and some of the early start-ups that had hoped to cash in on the move to blades have felt the pinch. FiberCycle Networks Inc., an early company created to sell low power blade servers quickly went out of business. Rival blade pioneer RLX Technologies Inc. has laid off staff and replaced senior executives.
"I don't think the vendors have made a compelling argument of what these things are good for," said Gartner Inc. analyst John Enck, who says that despite the blade vendors' best efforts to promote the blade concept as a simple step beyond rack-mounted computing, these new servers have yet to be accepted as general purpose machines.
One problem is pricing, Enck said. While blade systems have a price performance advantage when they are deployed in large systems, their costs are too high in configurations of less than 10 servers, he said. "These things are just too expensive," he observed. "There's no reason for blades to be premium priced, other than (the vendors) can get away with it."
Analysts also say that the complexity of blade computing -- with few standards in the areas of backplanes, management software, or even blade form factors -- is holding it back.
Not all users agree with the analysts, however.
"That's typical analyst stuff," said John Knuff, the vice president of network engineering with New York-based financial trading systems company NYFIX Inc. "I've never seen any new technology come out that didn't have a steep learning curve. It takes you a few months or a year to be good at it, just like anything else."
NYFIX is using blades to host high availability servers in its data centers that are running securities transactions, and Knuff is happy enough with the blade return on investment that he's planning use some more. "We're also going to use them for market data servers and we're starting to convert some of our existing Solaris applications to Linux and blade platforms."
Blades can be expensive, Knuff conceded. "If you've got inexpensive data center space, the blade server will cost you more than a regular server, as far as cost per rack," he said. But as the number of servers in play goes up, the economics of blade computing become more attractive. "When you're going to have 20 to 50 to hundreds of servers, then it's going to start to make sense," he said. "I think it's going to replace the 1U (1.75-inch-thick rack-mounted) server market, where people have one specific role for six or more servers," he said.
Others are less enthusiastic.
"We don't see anything really compelling about blades," said David Moffitt, the director of engineering at health care information technology company McKesson Corp. "The kind of environment that looks good for blades is when you have a lot of similar Web presentations that you need to expand and contract fairly rapidly."
Moffitt is not an ideal candidate for blade servers. His operation doesn't need to set up a large number of identically configured servers, and rack space is not at a premium at McKesson. With current blade pricing, and the nascent state of management tools, Moffitt remains unconvinced that the new form factor promises a compelling return on investment for his company.
The analyst consensus is not that blades will fail to take off, but rather that their acceptance will take more time than originally anticipated. IDC expects vendors to ship 137,000 units this year, accounting for $434 million in sales, and looking at numbers from the first two quarters of 2003, it appears that those targets will be hit, said IDC analyst Mark Melenovsky. "The market is certainly meeting our expectations now."
"It's amazing that the uptake has been as quick as it has," said IBM Director of eServer Product Marketing Jeff Benck. His company has shipped 17,000 blades since it released its BladeCenter product at the end of 2002, he said. He believes that blades will move beyond their niches in web caching and hosting to become a general purpose computing platform by the end of this year.
But whether the blade market will meet Benck's and IDC's expectations over the next few years is a matter of debate. Analyst firm Gartner Inc. expects the market to reach $1.2 billion by 2006, a respectable number, but far less than the $4.2 billion predicted by rival researcher IDC.
"Blades are going to happen. It really is inevitable," said Illuminata analyst Gordon Haff. But a combination of risk aversion on the part of buyers, and missteps by vendors has slowed down blade adoption, he said. These missteps include sellers bringing servers late to market or sending mixed signals about their product plans, and even technical problems with the products themselves, he said.
Dell's initial blades, for example, tended to overheat, and the company has since been "lukewarm" toward blades, Haff said. Blade servers are simply not mature enough to be considered the kind of commodity product that a company like Dell can successfully sell, he said. "It's basically not as comfortable a technology area for somebody like Dell," he said. "What Dell's really good at is churning out cookie cutter technology in large quantities."
But today's blade servers are not a cookie cutter technology. "Unlike rack-mount servers, and unlike towers, these things are just so radically different from each vendor that it's hard to do side-by-side comparisons," said Gartner's John Enck. "You'll see Dell come back in the market when there's more standardization."
Dell wouldn't say when it plans to ship a follow up to its PowerEdge 1655MC blade server, which was released last November, but other suppliers are moving ahead with new processor designs. Sun expects to ship AMD Athlon XP-M blade servers by year's end, and IBM plans to announce a blade based on its Power microprocessor architecture within the next three months.
The vendors disagree with the analysts on whether or not standardization of the blade form factor remains an issue. "Customers are asking for standardization of some of the management APIs and protocols,' said HP Vice President of Industry Standard Servers, Brad Anderson. "But standardization of form factors, I'm not hearing that at all."
"The form factor is basically a red herring because the customer manages the rack," said Neil Knox, Sun's executive vice president of volume systems. "Telecom tried to go to a single form factor, and they found that quite limiting."
Knox, like all of the blade vendors, says he's pleased with the rate of blade adoption, and confident that the new architecture will eventually find its place in this post-boom world. "It's never going to replace traditional servers; it's going to be a complementary architecture," he said. "Customers are no longer willing to rip and replace architectures. You have to coexist with the existing infrastructure."