Cutting to the Core Issues About Blades - Systems Management News On The Web

Cutting to the Core Issues About Blades
By Geoff Koch

August 15, 2008 — “Think of it in these terms,” says Steve Hill, an avuncular Wisconsin-based technology writer, consultant and frequent speaker on the blade server conference circuit. “Grab your wife’s hair dryer, aim it at yourself and turn it on at full temperature, then imagine five more blowing at you.”

Hill, an avid Green Bay Packer fan who the frequents the frozen tundra of Lambeau Field during football season, is talking about the other side of the temperature scale—extreme heat, a phenomenon he says is blade technology’s Achilles' heel

Assuming each runs at about 1kW, those six hair dryers going all at once, the resulting heat would be about as much as is generated in a typical rack and tower server. That’s heat, of course, that most data center managers have, through the years, learned to deal with by cobbling together approaches to cooling the server room, which in many businesses is one big, mostly homogeneously chilled space.

“Now, with blades, because the density is increasing, all of a sudden you can get 25kW in a rack,” says Hill. “And most data centers aren’t really designed to be able to deal with that kind of heat.”

Hill’s observation throws cold water on the triumphant claims coming from the big blade vendors, all of which readily point out that the ultra-thin slices of modular computing power are by far the hottest segment of a mostly stagnant global server market. Gartner Research predicts that, in terms of units shipped, the blade server market will grow at a compound annual rate of 19 percent from 2007 to 2012, pacing the overall market, which is slated to grow at a paltry 5 percent during the same time frame.

However, that same report warns that “dramatic growth does not translate to market domination,” forecasting that, despite these impressive sounding gains, by 2012 blades will still only account for one in five server shipments worldwide.

At the roughest cut, the first step in deciding whether to jump into the good news/bad news blade market is to look at a well-worn lists pros and cons, and decide which side of the ledger elicits the strongest gut response. Are you squeezed for space, drowning in cables and ever fearful of the next server outage? Or are you wary of vendor lock-in, content to let others work out the inevitable kinks associated with still-new technology, and happy (or stuck) with the current physical configuration and cooling capabilities of your data center?

Heat: The Downside of High Density
In the larger war between the big server vendors, one battle that IBM almost certainly has won is for the most memorable series of data center-related ads sprinkled across television and the Web featuring Gil, the hapless, slightly overweight hero.

In one, Gil is seen sweating and stumbling through a parched desert strewn with racks of servers. “The servers are too hot! Save me!” he screams before collapsing in what is apparently a dream. In the next frame, Gil finds himself awake on the floor of a sterile, spacious server room. “I already have,” says Gil’s young, hip counterpart. “I got an IBM BladeCenter,” adding, as he slides a blade into a chassis. “Runs cooler, saves energy.”

The ad (search for “Out with Heat. In with Blades” on YouTube), though memorable, is also misleading. It’s true that compared to various rack-mounted alternatives, blades consume as much as 35 percent less power for an equivalent amount of computing horsepower. However, this power and equivalent heat is more condensed, leading to hyper-local hotspots, which can be problematic for data centers stuck with aging cooling systems.

Configurations vary, but generally these cooling systems are comprised of some combination of vented raised flooring with various computer room air conditioners. These CRACs, which blast icy air under the raised floor and up through the vents, may not be sufficient to deal with heat challenges of blades.

“Airflow is another key aspect of cooling servers, measured in cubic-feet-per-minute (cfm),” write consultants Barb Goldworm and Anne Skamarock in their 2007 book "Blade Servers and Virtualization."

Citing data from power and cooling vendor APC, the pair note that blade server systems typically need 120 cfm of cool air per kilowatt of power. This is problematic since most data centers provide 200-300 cfm of cool air per rack location, a measly 10 percent of what is required.

“If no additional cooling is used,” Goldworm and Skamarock write, “this would limit rack power to less than 2kW per rack, and blades would not be an option.”

The solution, which can include reconfiguring the data center, ripping up floors and even buying supplemental pumped-water or refrigerant-based systems, may at first blush sound too expensive for many businesses, particularly in a sagging economy in which IT budgets are being squeezed.

But scrimping now may turn out to be penny wise and pound foolish. Most vendors can cite migraine-causing statistics about the time it takes for power and cooling costs to outstrip the initial purchase price of an average server. Hewlett-Packard, current market-share leader for blade servers, pegs the number at around three years.

Then there’s a study released Apr. 30 by McKinsey & Co., which found that the world’s data centers will pass the airline industry as greenhouse gas polluters by 2020. Released at the Green Enterprise Computing Symposium in Orlando, the study found that computer servers are used at only 6 percent of their capacity on average, while data center facilities as a whole are used at just 56 percent of peak performance.

“In other words,” wrote New York Times tech reporter Steve Lohr in a May 1 posting to his Bits blog, “if data centers were hotels, they would be bankrupt and shut down instead of growing like kudzu.”

The McKinsey study suggested corporations should set the goal of doubling the efficiency of their data centers by 2012, proposing a new metric called CADE, for Corporate Average Data Efficiency, purposefully alluding to the government’s fuel efficiency (CAFE) standards for cars.

And the government itself has gotten into the act, mandating the Environmental Protection Agency to prepare a lengthy report on data center energy usage as part of Public Law 109-431, passed in December 2006. The August 2007 EPA report found that “Unless energy efficiency is improved beyond current trends, the federal government’s electricity cost for servers and data centers could be nearly US$740 million annually by 2011.”

With sustained high prices at the pump and both presidential candidates on the record in support of energy policies that help reduce carbon emissions, it’s not a stretch to suggest that someday soon, something like blade-enabled efficiency might be more of a government mandate than a voluntary response to a vendor sales pitch.

Cure for the Cabling Blues
For now though, it’s a fair bet that most IT managers are more fearful of unruly and expensive tangles of cabling than unwelcome government interference. Indeed, perhaps the most iconic image in blade-related literature, advertising and otherwise, is the mess of multicolored cables sprawling from the rear of a rack server. The caption in Goldworm and Skamarock’s book under one such image reads, “A server-room rack-server cabling nightmare.” And yes, there is another ad starring Gil. (See “Out With Cables. In With Blades” at YouTube.)

Dealing with this inevitable Gordian Knot of Ethernet and Fibre Channel cable is one reason that many companies have to wait for scheduled downtime to swap out a server.

“Somewhere along the line, something has gone wrong with how we do data centers,” says Houston-based Jim Ganthier, director of marketing for HP’s blade product lines. “And what I mean by that is it takes three, maybe even four people to get a server online.

“You’ve got to have a facilities person, a SAN admin, a server admin and a LAN admin,” he continues. “They all get together, they trade a bunch of information in multiple meetings; we basically said, ‘Stop the insanity.’ ”

Ganthier is referring to HP’s Virtual Connect technology, which allows customers to pre-assign network and storage connections, thus saving time when deploying servers. He says the technology, which supports the company’s BladeSystem c-Class products and has only been available since February 2007, is already used by thousands of customers.

Vendor-specific products aside, most blade product lines allow for lots of consolidation around most interconnects, including Fibre Channel, which is predominantly used today to connect servers to large storage systems, especially SANs.

For rack-mounted and other commodity-type servers, it’s not uncommon for each server to be connected to the SAN with a Fibre Channel adapter and associated cabling, which can be relatively expensive (adapters alone can top $1,000 apiece). Most blade systems allow for aggregating the Fibre Channel connections at the backplane, reducing the number of links to individual servers.

“If you start looking at the numbers in terms of savings of energy, savings of redundancy of gear and then simplicity of managing them—you know, being able to control 10 servers via one standardized interface—all of a sudden the numbers start to add up,” says Hill, author of an exhaustive late-2006 feature article, product review and research report on blades for “Network Computing.” “But they add up for big customers and not so much for the smaller ones.”

Look Out for Lock-In
One reality that all potential blade customers face is a certain amount of vendor lock-in that comes with every blade purchase. Almost without exception, you simply can’t plug a blade server from one vendor into the chassis of another.

All of the vendors are quick to wave away any concerns about proprietary lock-in and to point to all the ways they support standardization, such as their support of a variety of switching and other I/O-related technology. IBM and HP, which together account for the lion’s share of the blade market, both insist that it’s a non-issue for customers and that it’s unlikely that any significant chassis interoperability efforts will be forthcoming in the near future.

“Yeah, that’s a-nice-to-have, but truthfully, I don’t expect to be able to take the tires off my Suburban and put them on my wife’s BMW,” says Ganthier.

“Sure, the customers do ask for more open systems and the second-tier players also want to break the barriers,” says Matthew Wilkins, a U.K.-based analyst for iSuppli. “But the major OEMs do not want share [chassis specs] with their competitors; they are in good positions already and proprietary components could keep their competitors away.”

The one entirely modest exception to the general rule of electro-mechanical exclusion comes out of Blade.org, an IBM-initiated effort to accelerate the development and adoption of blade server solutions, mostly around the company’s BladeCenter products.

As a carrot to encourage participation in the consortium, IBM is sharing some of the details of its BladeCenter architecture. One result is a recent announcement that Themis, a small Fremont, Calif.-based specialty computer maker, has used the specification to build a blade running Sun’s UltraSPARC processor for IBM BladeCenter.

“This thing runs SPARC/Solaris applications, right inside our chassis, right beside all our traditional IBM stuff,” says Scott Tease, product marketing manager for IBM’s BladeCenter products. “I think that kind of thinking may get you to the kind of openness that clients are asking for, and this is an area in which we are way out ahead of our competitors.”

Legacy Thinking
As Tease describes it, one of the reasons behind relatively low blade adoption is predictable half-a-step-behind thinking, something that plagues most new technologies. Customers, he says, still think of blades as most useful for applications that would require 100- to 1,000-node Linux clusters.

“People used to buy blades because they were super dense and high performance—almost always for a big data center play,” he says. “But we’re now seeing tremendous interest around blades for the smaller client, the non-data-center-owner.”

IBM, which launched last summer its BladeCenter S products for small- and medium-sized businesses, says that segment now accounts for around 20 percent of its overall BladeCenter business. “It’s a big growth area for us,” adds Tease. “A year and a half from now, it could easily be 40 percent.”

Goldworm and Skamarock declare blades, along with their software counterpart in virtualization, to be part of an inevitable evolutionary step in computing, one as significant as the commoditization of the PC and the popularization of the Internet. “Blades and virtualization are key enablers of true utility computing,” they write, “An environment in which components are added (provisioned) when they are needed, where they are needed and only for as long as they are needed—to give control back to businesses.”

But of course, control is wielded in multiple ways in technology. While hotshot 20-something uber-geeks may flock to the cool design and programming jobs associated with gaming or various Web 2.0 technologies, the task of running data centers is still likely mostly in the hands of grownups, many of whom have become quite comfortable with their server racks, cables and all.

Hill says it often takes a fresh perspective and new blood to lead to change, no matter the technical or cost advantages of a new technology.

“It’s the new guys who are coming in who say, ‘Holy crap, why aren’t we doing this? I could put four chassis in here and never have to mess with it,’ ” says Hill. “Recently, I was talking to an attendee of a conference I’d been speaking at, and he came up to me laughing and said, ‘You know, when I first joined this company I was told there was going to be no blade servers and no virtualization. Today, a large majority of our applications are virtualized on blade servers—and that guy is now my peer instead of my boss.’ ”

Blades by the Numbers
Analysts agree that blades, despite relatively modest market penetration, will be the hottest-selling server offering well into the near future.

Gartner Research and iSuppli are mostly in sync on their 2008 forecasts for blades: both firms expect blades to account for roughly 1.2 million units out of the overall 9.4 million servers expected to ship worldwide this year.

And both are bullish on the future: iSuppli predicts a sizzling 31.5 percent compound annual growth rate from 2006 to 2011 while Gartner, whose most recent forecast covers 2008 to 2012, sees a less fulsome growth rate of 18.6 percent.

Among Gartner’s other predictions in its June 16 report: “Blades and chassis will not be standardized across vendors in the five-year forecast.”

If you’re a newbie—and based on current market share data, 90 percent of you are—then you need a good reference book that goes beyond the slanted whitepapers, case studies and spec sheets you’ll find on vendors’ Web sites.

“Blade Servers and Virtualization,” by Barb Goldworm and Anne Skamarock is a good place to start, if for no other reason than to peruse its collection of practical lists and tips.

The five most-pressing questions for any blade-leaning data center manager to ask:

»	Do you have requirements for a large number of network connections per server (more than eight)?
»	*Do you have requirements for specific chipsets, or for unique processing or graphics applications, which may or may not be available in a blade form factor for your chassis?
»	Do you require a very large number of local hard drives for each server?
»	Are you addressing environments with four or fewer physical servers in one location?
»	Is your data center reaching the power and cooling limitations of the facility? If so, would the blade servers augment or replace the existing legacy server hardware (through consolidation—either physical, virtual or both)?

“If the answer to all these questions is no,” Goldworm and Skamarock write, “Then nothing should be stopping you from gaining the many benefits of blade systems.”

Back in 2005, with 35 percent of blade revenue, HP was firmly lodged in the No. 2 spot behind IBM. But the Palo Alto, Calif.-based company, which in July announced that it was the first blade vendor to pass the one million units shipped milestone, now is the market leader, commanding about half of the worldwide market, according to both Gartner Research and IDC.

Credit its BladeSystem c-Class product line with the growth, along with a commitment to be “as pragmatic as hell,” says HP's Ganthier.

Will HP consider joining Blade.org, the IBM-sponsored consortium attempting to bring something of the open-source ethic to the blade market?

“Why would I want to be part of Blade.org, when we have three times more partners with HP BladeSystem Solution Builder program?” he says, adding that expecting chassis-level interoperability in the blade market is flat out unrealistic. “Eli Whitney is a great aspirational goal, but it’s just not something you see in many industries or markets.”

Related Search Term(s): blades, HVAC, power, server hardware, virtualization, HP, IBM

Share this link: http://www.sysmannews.com/link/32727