eWEEK content and product recommendations are editorially independent. We may make money when you click on links to our partners. Learn More.
Hewlett-Packard will roll out a more powerful and efficient fan design in summer 2006 for its blade servers that was inspired by devices used in some remote-controlled model airplanes.
HPs Active Cool Fans design will first appear in the next generation of the Palo Alto, Calif., companys BladeSystem blade servers, said Paul Perez, vice president of storage, networking and infrastructure for HPs Industry Standard Servers unit.
After that, HP will look for other devices in the data center that could use the new design, Perez said.
He declined to elaborate, though he did say that company engineers already know how the design could be implemented in other machines.
The Active Cool Fans design, which was introduced May 3 at a breakfast event in San Francisco with West Coast analysts and reporters, is one of several initiatives under way at HP to address the twin issues of power consumption and heat generation in data centers.
The development of more dense servers and more powerful processors, coupled with increased electricity rates, has helped fuel a rise in power costs for businesses, both in the running of the systems and the cooling infrastructures inside the data centers.
“There is a lot of emphasis within the industry to address the power issues,” Perez said during an interview after the event.
The new fan design uses an electrical ducted fan, or EDF, approach, he said. Currently, air from outside the server rack is sucked in via fans and blown across the inside of the systems to keep the components cooled, and then blown out the other side of the rack.
In an EDF design, air is brought in, then directed through sealed ducts to components that need it most.
Perez said the result is a system that is higher performing but more efficient, both in its cooling capacity and the amount of power thats needed.
The new design offers four times the performance of current fan systems. At the same time, a server will use a third of the power to run those fans.
A typical 1U (1.75-inch) rack server with an average of 18 fans uses 60 watts to spin those fans, Perez said. The new design will use about 20 watts on average, he said.
The seed for the new fan design was planted about three or four years ago, when Ron Noblett, vice president for shared engineering services for ISS and a remote-controlled airplane enthusiast, saw that people who built models of jets used an EDF system.
He asked Wade Vinson, a thermal engineer within HP, to see how that system could be applied to what HP was doing with power and cooling in its servers.
The result was a design that garnered 20 patents, Perez said—10 for the fan design and another 10 for the usage models of those fans.
The Active Cool Fans design will enable HP to enhance the density capabilities of the blade servers when the next generation is released this summer, an important step in increasing adoption, Perez said.
Blades have offered the promise of greater computing power in small spaces, but the power and cooling issues have hampered adoption, according to industry observers.
In the third quarter in 2005, IBM held 42 percent of the blade market, followed by HP at 32 percent and Dell at 9 percent, according to research firm IDC.
“While we were in the process of designing new blade systems … we needed to do something different with both power and cooling,” Perez said.
“So we looked at doing something different. We didnt want it to be evolutionary. We wanted it to be disruptive, to be revolutionary.”
A Rallying Point
The power issue has become a rallying point for the industry.
On April 19, HP and a number of other vendors, including IBM, Sun Microsystems and Advanced Micro Devices formed the Green Grid Alliance, which is working to make corporate data centers more energy efficient.
At an event in New York on May 2, several other companies, including Dell, American Power Conversion and VMware, also joined the alliance.
Vendors also are trying to make their individual products more efficient, both through hardware designs and management software.
Both HP and IBM have introduced liquid-cooling devices, and Suns UltraSPARC T1 chip offers up to eight cores in a 60-watt package.
AMD and Intel also are working to make their processors more efficient.
Perez said HP is working on several other fronts to make its systems more efficient.
HP engineers are looking to use “pervasive sensing” capabilities to enable systems to dynamically speed up or down individual fans within the server, depending on the amount of heat in the area of the fans.
At the same time, the company is looking to use the same capabilities within the data center, where the flow of coolant and fans in data center air-conditioning units can be dynamically managed depending on where the most heat is found.
“Were doing this at the macro level in the data center and the micro level in the blade enclosures,” Perez said.
In addition, HP is looking to bring “dynamic capping” capabilities to its blade systems.
In the current “static capping” scenario, users can put a cap on the amount of power a rack consumes—say 8 or 10 watts—and each server is capped at the same rate to reach that limit.
With dynamic capping, individual blades can be powered up and down depending on the workloads in each, all within the rack power limit set by the user.
HP also has proposed to the SPEC (Standard Performance Evaluation Corp.) standards group metrics that can be used to compare the power efficiency of systems.
Power consumption is a growing concern among CIOs, but there is no standard way of measuring—and thus, comparing—the power consumption of different machines, unlike the automobile industry, where cars can be measured by the miles per gallon they can get.
In a metric submitted to the SPEC group, HP proposes a system where consumption is measured on a server at different utilization levels—at 20, 40, 60 and 100 percent—with the average at each level used as a benchmark.
Perez said its a way of “rating different servers on the same workloads.”
He said he expects the proposal to go through a three-to-six-month review process before anything is finalized.