IBM has designed a new type of rack-mount server specifically for companies running heavily trafficked Web 2.0 sites such as Facebook and MySpace, the company announced Wednesday.
Called the iDataPlex, the server is designed to compete with the unbranded "white-box" PCs that online companies link together by the thousand to run busy Web sites. IBM said its new server, which runs Linux and is based on Intel's quad-core Xeon processors, consumes 40 percent less power and packs more computing punch than a typical rack-mount system. The energy savings come largely from a new design that requires less power for cooling, IBM said.
Rack servers are the slender machines shaped like an oblong pizza box and stacked on top of each other in server chassis. The servers come in standard heights -- 1u or 2u -- but their depth, or how far back they reach into the chassis, has been expanding as vendors try to cram more hardware components inside.
That has created a problem, according to IBM. Cooling systems blow air over the servers from back to front, and as the servers become deeper it takes more energy to power the fans that cool them. "The power used by the fan is proportional to the cube of the fan speed, so if you want to double the fan speed you have to use eight times the power," said Gregg McKnight, CTO of IBM's modular systems group.
IBM's answer is to rotate the server horizontally through 90 degrees, producing a server that is wider than usual but only 15 inches deep, compared to about 25 inches for a typical rack server. "That allowed us to run fans at a much lower velocity, and therefore save about 67 percent on the fan energy alone," he said.
IBM also pushed two racks together, creating a single wide rack that holds 84 iDataPlex servers. That allowed it to share three power whips between the servers, while two separate racks would normally use four. Power whips, the moveable outlets attached to power cables, cost $1,500 to $2,000 per month to maintain, McKnight said.
The broad surface area at the back also allowed IBM to design an optional water-cooled rear-door heat exchanger, which IBM said extracts all of the heat from the system, so it doesn't contribute to datacenter warming.
The trade-off for sharing power cables is a less fault-tolerant system, but the software used to run busy Web sites is usually designed to fail over quickly to another server. "We interviewed Web 2.0 companies and they told us unanimously that they are designing their applications to tolerate server failures. So because it's more economical and more energy efficient, it's an attractive trade-off for them," McKnight said.
Joe Clabby, president of Clabby Analytics, in Yarmouth, Maine, said that philosophy marks a significant change. The big server vendors have focussed on designing costly machines with high levels of redundancy, which is what companies need to run back-office business applications where every transaction is critical.
"In the Web 2.0 world, you don't need that," Clabby said. "If an IP address drops out, they just kill the server and move to another one. They're not focussed on business resiliency here, they're focussed on fast, cheap and fix-it-on-the-fly. That's a big change for IBM and the server industry."