It’s a long way from Tempe, Ariz., to Mars. So when distant and expensive data comes streaming in from NASA satellites orbiting the red planet and arrives at Arizona State University’s Mars Space Flight Facility, it’s important to utilize the information in the most cost-effective way. That’s why the lab started using blade servers. “More CPUs are definitely better, and the cluster approach provides greater than 10 times more CPUs than, say, a Sun symmetric multiprocessing machine at the same cost,” says Noel Gorelick, the lab’s software manager.
Blade servers are compact because blades?processor, memory and storage on a card?reside in rack-mountable enclosures that supply power, ventilation and other support components. Like most cluster-based systems, blade servers can be configured to include load-balancing and failover (another blade can automatically step in to take the place of a failed card). Individual blades are usually hot-pluggable, which makes it easy to swap out a board with a new one in the event of system failure (for more details, see “Servers on the Edge,” www.cio.com/printlinks).
As enterprises begin to recognize blade servers’ potential to save space and cut costs, the technology has become increasingly popular. More than 150,000 individual blades were shipped in 2003, says Jeffrey J. Hewitt, a principal analyst at Gartner. That’s a big jump from the year before when just 30,000 units shipped, he says. And market researcher IDC (a sister company to CIO’s publisher) predicts that 2.2 million blades will ship worldwide by 2007, representing 27 percent of all new servers sold.
Mission for Mars
The Mars Space Flight Facility at Arizona State University (ASU) is using a BladeRack system from RackSaver. The Mars team installed the system in January 2003 and uses it to translate the thermal emission imaging system data, gathered by instruments on satellites orbiting Mars, into clean, detailed images that are suitable for study. The facility’s images helped NASA find optimal landing spots for the January 2004 landings of the Mars rovers.
Low cost, reliability and raw computational power are important to a program that performs cutting-edge science under a tight budget. “We usually build most of our computers in-house,” says Gorelick. “We elected to buy the BladeRack because the vertical profile of each blade takes less space than an equivalent product we could build. Also, the community cooling, power and networking remove several per-machine failure points.”
The biggest problem the Mars team experienced with its blade server was moving it into the building. “We were unable to get it into the Space Flight Facility without laying it on its side,” says Gorelick. To reduce the unit’s weight, allowing it to be safely tipped over, all of the blades had to be removed. The handling took its toll on the blades, however. “To no one’s surprise, several ended up DOA, and several more died within the first month,” says Gorelick. The blades were quickly replaced, though, and the system has since operated without any major problems.
The Mars researchers are so impressed with their blade server that they want to extend the technology to other ASU technical projects. “We’re hoping to pool research money from several computer-intensive projects to produce a center for scientific computing,” says Gorelick. “That could involve as many as 1,000 more nodes.”
On a Wing and a Blade
Closer to Earth, Continental Airlines is relying on blade server technology to power its Web presence. Today, the company uses 24 Hewlett-Packard ProLiant BL40p four-processor blades, running Windows 2000, as Web servers. The Intel Xeon-processor blades are linked to conventional ProLiant servers that function as database clusters, giving customers the ability to view, price and book flights, as well as offering the airline easily scalable performance, space savings and streamlined management. The configuration is set to double in size when a backup Web presence goes online later this year.
The blade servers replaced several older Pentium-based ProLiant 580 servers, says Bob Edwards, vice president of systems technology for Continental Airlines. “We needed more power to handle Continental.com’s sales, and when we did the upgrade we just transitioned over to blade technology.”
Like most other blade server adopters, Continental acquired the technology to boost performance and save space, which should reduce costs in the long run. The system also helps the airline quickly scale its website to suit business surges and lulls. With the old technology, getting a new server online could take the better part of a day. “It was too late; the sale was over,” says Edwards. Blade servers allow Continental to respond to rising demand much faster. “We can now deploy an image onto a Web server?application and OS?in probably one-tenth the time,” says Edwards.
The blade servers haven’t given Edwards any major headaches so far. He does, however, worry about the system’s physical environment. “Because of the density, there’s a lot of heat generated,” he says. Like many other blade adopters, Edwards faces the prospect of adding enhanced cooling, such as improving the airflow between enclosures or even adding a water cooling system. “I did not anticipate the heat distribution problem,” says Edwards. “You need to pay careful attention to where the servers are placed and how heat is managed in and around [them].”
On the other hand, Edwards appreciates the technology’s space-saving attributes. “We had five racks originally; we’re down to three racks now. That’s a pretty substantial reduction in floor space,” he notes. And deployment took only one month.
Compared to conventional servers, managing blades is a snap, says Edwards. He notes that having the servers near each other and managing them with HP’s Insight Manager software streamlines administration. “For example, where we have 48 servers, the management utilities allow you to manage six servers in one array,” says Edwards. As a result, the system has cut in half the time required by system administration duties. Edwards says this capability will free up staff to handle other tasks.
Although he’s now a blade server believer, Edwards says that as recently as a year ago he wouldn’t have considered adopting the technology because of concerns that blade servers hadn’t yet proved their value. But time changed his mind. “We just felt like the technology had matured, and the advantages of using it for this particular application outweighed the disadvantages and risks.”
While sales are rising, blade server prices are heading in the opposite direction, thanks to strong competition between Dell, Hewlett-Packard, IBM, RackSaver, RLX Technologies, Sun Microsystems and other blade server vendors. Dell’s blades, for example, start at about $1,499; enclosures begin at $1,799. Although ultracompact blade servers are significantly more expensive up front than their standalone counterparts (IBM’s Xeon-powered BladeCenter HS20 costs around $2,500 with software, versus $1,400 for an equivalent standalone IBM xSeries 225), enterprises can reap rewards in reduced floor space, lower power consumption, simplified network cabling, and easier maintenance and administration.
Beyond the technology and attractive prices, enterprises are being drawn toward blades because the products are establishing a favorable track record. “People have had enough experience, at least the early adopters have, that they can see examples of how blades can really work effectively to help cut costs,” says Rebecca Wettemann, vice president of research at technology analysis company Nucleus Research.
Upcoming “diskless blades”?units with no hard drive?promise to make blade servers even more popular. These devices increase the servers’ mean time between failures, since drives tend to be the major source of hardware disasters. The technology will also let enterprises fully leverage a storage infrastructure?including storage area networks?into the blade environment. “Diskless blades are configured to be able to boot from external storage and utilize all of their disk storage needs using external devices rather than accessing local drives,” says Gartner’s Hewitt. “It will take some technical efforts to achieve this and have the blade servers run efficiently.” Gartner predicts diskless blades will be available from all major server vendors by the end of 2008, although many blade server makers will offer the technology well before then.
The pending arrival of grid computing should also help boost blade server sales. Many see blades as a natural fit with grid technology, given blades’ flexibility, high performance and cluster-based foundation (see “True Grid,” Page 44).
With blade servers now becoming entrenched in the IT mainstream, adopters and vendors are beginning to face issues that often affect maturing technologies. With blade servers, two key concerns are standardization and improved server management.
Continental’s Edwards is pessimistic that there will ever be meaningful blade server standardization, given the current cutthroat competition that exists between blade vendors. “The idea of being able to take any vendor’s product and plug it into that rack seems far-fetched to me,” he says. Because of this, Edwards suggests that enterprises buy blade server technology to address a specific, immediate application. “Just like you would for a midrange system,” he says. “Put it on the data center floor and plan on replacing the whole thing when it comes time to upgrade.”
Blade management tools are often supplied to customers by the hardware vendors themselves. IBM, for example, bundles its Director management software free with its eServer blades. This means that many customers are drawn toward a particular blade server vendor on the basis of the company’s management tools as well as its hardware. Software quality varies widely between vendors, so it pays to shop carefully to find the right tools. “If [the vendor] gets the management right, a blade server approach can deliver a great return,” says Wettemann.
As the blade server market matures, more third-party software companies are entering the field, which is helping to expand the pool of available products. ASU’s Mars team, for example, relies on job scheduling software from PBS Pro. Companies such as BladeLogic and Opsware also offer automation software to manage blades. These companies provide products that automate many configuration processes that are usually handled manually, such as the application of patches or the collection of inventory information. One advantage these companies offer is that their software can work with blade servers and traditional servers from a variety of vendors. Hewitt believes that blade management tools are destined to get better and become more widely available in the near future. “It will probably take two years to show maturity,” he says.
In the meantime, for most adopters, blade servers’ shortcomings in the areas of cooling, limited software selection, hardware incompatibility and limited track records are a small price to pay for powerful server capabilities in a small package at a lower cost. As Gorelick puts it, “More CPUs are better.”