By SUSAN LAHEY
Special contributor to Silicon Hills News
Existing data centers–often giant buildings filled with servers that store, analyze and process information for individual companies, the internet and cloud computing systems–pump out tremendous amounts of heat but require cooling to protect the data. Globally they consumed 31 gigawatts of energy last year—roughly the same amount as New York City consumed on its highest peak day back in 2006. Evans’ idea is to switch from giant, sweaty workhorse servers to small, efficient servers on tiny batteries.
At first people tell him he’s crazy. Not only that, but the recession causes all venture capital funds to dry up for early startups and it looks like Evans has a much longer row to hoe than he anticipated. But he and his partners continue to experiment with the idea and two years later, recession or not, his company raises $48 million in venture financing—a huge sum for a startup that doesn’t intend to manufacture its own product. A year after that Calxeda announces who will be handling the server side of its venture: server giant Hewlett Packard.
“Companies were building data centers by rivers, in Finland…everybody was working to solve the issues around the problem but nobody was working to solve the problem,” said Larry Wikelius, Calxeda cofounder and vice president of software engineering . “This is the first time somebody said, ‘We can take power out of the equation.’”
Evans is a 20-year veteran of the semiconductor industry who started Calxeda after leaving his position as vice president and general manager of Marvell’s Application Processing Business Unit in the Cellular and Handheld Group (acquired from Intel). He spent a year talking to end users about using a mobile-type power source for servers, one that would only require a tenth the energy of existing servers and would work better with servers fluctuating energy needs.
His first partners were his cofounders, Wikelius and David Borland, vice president of hardware engineering. The company planned to model the chip somewhat after those being produced by British mobile chip manufacturer, ARM, and made several attempts before figuring out how to do it. It was crucial to the founders to create a solution that could be plugged into existing data centers without major overhauls, Wikelius said.
“Too many companies say ‘If you just remove your software and change all your infrastructure we have a great product for you,’” he said. “We support standards so it’s easy to plug in to the existing infrastructure.”
But it’s not exactly plug and play, according to Fichera.
“ If you mean recompile, reload and it works fine, with some applications it works that way,” he said. “Others will have to be completely recoded.”
Still, Borland said, finding the simplest solution is part of Calxeda culture. “Calxeda is a loose Latin translation of the name we started with—Smooth Stone. “ In the David and Goliath story, he points out, when everybody else despaired in the face of a big problem, David applied the simplest answer, a smooth stone.
“We’re not interested in science projects, ideas that make you say ‘That’s really cool, but how do you apply it” Wikelius said.
Finding the simplest solution, however, has required building a complex network of partners who can speak to all the development issues. And that, too, has become part of Calxeda’s culture. The company has a dozen or so partners it talks about in its Trailblazer initiative. But it has many more it doesn’t talk about. One it has connected with is Pervasive Data Rush, a big data solution of Pervasive Software.
“Calxeda and Pervasive have a shared interest in providing solutions for one of the key challenges of IT today – the economics and energy consumption of data centers,” said David Inbar, senior director of sales and marketing of Pervasive Big Data Products and Solutions. “The problem is, typically only 15 to 30 percent of processor capacity is being utilized at any given time. The rest of the capacity is idle, but still burning energy. Pervasive DataRush technology gives people the ability to build programs that run in parallel, taking full advantage of processors to get lower costs and better energy efficiency.”
That’s the kind of problem Calxeda’s been hashing over with its partners to make sure its solutions work. The founders were deliberate about the order in which they moved forward, building the company for three years before announcing its server partner. They kept their options open as long as they could, Wikelius said, because things change fast in this industry and you may have knowledge and information next year that’s missing this year.
When it did finally announce its server manufacturing partner November first, it put the little Austin startup in a whole new class. To come out of the gate with the most prominent dance partner on the floor sets the company well ahead of competitors.
And it does have competitors. Other companies in the game include Applied Micro which has an intermediate level ARM based chip for servers. But, said Fichera, Calxeda’s offering a lot of different pieces of the architecture on one server which sets them ahead in reducing the footprint and the need for power. Calxeda has other advantages. One was Evan’s experience in the mobile space. Another, Wikelius said, was the company’s intense focus on producing this product for internet and cloud server industries. Big companies couldn’t afford such focus. And finally, Evans said: “You just have to go faster than everybody else.”
Teaming up with HP puts Calxeda on a new playing field. HPs new servers will be called Redstone. They’ll take up a fraction of the space a rack of servers formerly required and will share resources. As a Wall Street Journal story put it they will “sip energy lightly.” HP announced that a typical server project of 400 servers in 10 racks would cost $3.3 million to purchase, install and operate. Redstone runs four times as many servers in half of one rack and draws a tenth the energy, dropping the price to $1.2 million.
In the near future, Fichera said, the impact of Calxeda’s work will be seen more in the transformation of the way the industry thinks about data centers, rather than in big profits. After a few years, however, it’s likely to be lucrative niche.
In the meantime, Calxeda and its partners are thoroughly enjoying themselves. The variety of working with all these partners makes Calxeda a fascinating ecosystem.
“There are a bunch of people who would never work for the same company in our partnerships, it makes a whole ecosystem,” Borland said. “It makes us much more than a chip provider.”
From the partner perspective, Inbar said, “it’s fun (working with Calxeda). We share a lot of culture.”
The partnerships work for a couple reasons. For one, Wikelius said, Calxeda is very transparent with partners and employees. For another, they’ve created a culture that invites people to rethink everything. Just because a server has always been a giant box and a small chip runs mobile phones doesn’t mean it has to stay that way.
Working in that environment, with these people is Evans’ favorite part.
“I love it,” he said. “I’m enjoying the ride.”