The Crusoe processor happened along at the right time for Christopher Hipp, a graphic designer turned systems entrepreneur who had recently founded a server company, RLX Technologies Inc., in The Woodlands, Texas. Inspired by the efficiencies of early Cobalt rack configuration, Hipp and his venture capital partners anticipated a blade design built around Intel Corp.s mobile Pentiums or Celerons. But they quickly discovered that the heat from those chips prevented using a lot of them in close quarters.
“So, when Transmeta Corp. came along in early 2000 and announced a processor that was 85 percent to 90 percent of the mobile Pentiums performance with a fifth of the power consumption, it was a no-brainer,” Hipp said.
The result was the RLX System 324, a blade configuration that packs more punch into a smaller space than any other server on the market—up to 336 blades in a single, 42-unit, industry-standard rack, which is about eight times the processing power that can be squeezed into the same space using a 1U (1.75-inch), Intel-based, “pizza box” chassis.
While the Crusoe has a 10 percent to 15 percent performance disadvantage, there are jobs at which it cant be matched. For example, a fully populated RLX System 324 rack can serve 372,000 Web pages per second—a fivefold advantage over the 71,000 pages a corresponding Compaq Computer Corp.-based system with 42 1U servers can deliver, according to the RLX white paper “Redefining Server Technologies.”
“For an app server or Web server, Domain Name System, intrusion detection system, or mail, I think RLX is fantastic,” said Dwight Gibbs, an IT consultant who converted The Motley Fool Inc.s co-location space to RLX servers when he was the Herndon, Va., companys IT manager. Gibbs said he recently installed another RLX unit for a company that will be running an application server. “All I needed was two 120-volt circuits,” he said. “It was really easy.”
So why isnt every data center and IT shop replacing legacy systems with RLX- or other Crusoe-based servers?
“Were simply not a replacement for some of the high-end systems out there,” Hipp said. “Were very good at certain areas, and that is front-end Web applications and other areas where applications can be highly distributed, but we are not a replacement for a high-end SMP [symmetric multiprocessing] database machine.”
Sometimes, what does work surprises even Hipp. For example, a group at power-starved Los Alamos National Laboratories, in Los Alamos, N.M., recently said it had found a way to squeeze supercomputer performance out of an innovative configuration of RLX servers.
Hipps biggest problem right now is a lack of Crusoe processors. Last fall, Transmeta, in Santa Clara, Calif., tried to cut costs by switching fabrication operations from IBM, which had produced the chip since its introduction, to an overseas plant. The result has been a disaster for Transmeta and RLX, as quality control problems proved so extreme that Transmeta had to suspend shipments. They are expected to resume next month at the earliest.