10 innovators changing the game for Internet infrastructure

The Wildcard: Barry Evans, CEO, Calxeda

by Stacey Higginbotham

The chip world is full of outsized personalities. Think of Jerry Sanders and his declaration “Real men have fabs,” or Nvidia’s CEO Jen-Hsun Huang, who has gotten into a number of verbal tiffs with rival companies. So when Barry Evans, the CEO of a startup that aims to take on Nvidia (s NVDA), Intel (s INTC), Marvel (s MVL) and other established vendors, walks into the room, you’d think you’d notice.

But Evans, who founded and heads Calxeda, a company building a server that will use the same chips found in cell phones, isn’t given to outsized pronouncements or even a lot of laughs. He’s so unassuming that his coworkers have made him a sock puppet (They call it SoC Man, short for system on a chip) that sits in his office with a sign that says, “I’m excited,” which he can use to indicate that, yes, he is indeed excited.

Despite his calm exterior, he is passionate about Calxeda and his vision that the world of computing has changed and the chips and gear sold to webscale and cloud operators will have to change with it. Computing has become too important and too big to rely on chips that can consume 20 to 60 watts of power, especially since many of those machines are virtualized and could theoretically cut performance and still perform the jobs they are asked to do.

Calxeda is building servers that use the low-power ARM architecture — sacrificing performance for power consumption. Calxeda’s first product is a 5-watt server that delivers many ARM cores running at between 1.1 GHz and 1.4 GHz. HP is working with Calxeda to deliver a complete system, while other hardware vendors including Dell are experimenting with ARM-based systems using chips made by other companies.

The growing ARM ecosystem is a testament to long-range planning by folks like Evans. Unlike software, silicon has to be designed years in advance of ever running in a production server. Evans says back in 2007, when he was working at Marvell (he ended up there after Marvell bought the ARM-based chip business from Intel, where Evans had worked), he started seeing the scale of Google’s and Amazon’s operations and saw the EPA’s projections about power use in the data center. He realized things couldn’t continue in this trend.

As the former head of the ARM-based X-scale business at Intel (and then at Marvell) Evans wondered if a new architecture for servers was the answer. “You know when you have something and you feel it could be an awesome idea, and then a few years later you see it elsewhere and say to yourself, ‘I had that idea back in whatever,’ and you kick yourself for not going after it?” he asks. “This one felt really big, and so I had to go after it.”

So Evans started doing the research and wrote a business plan, and then went to ARM. “Their response was ‘What’s a server?’“ Evans said. But he ultimately convinced ARM and then managed to raise $48 million from venture investors for Calxeda.

As the world of IT changes, the concept of a data center as a computer or even software defined infrastructure is one that can play to his advantage. One of the biggest impediments to running an ARM-based server is that most of the enterprise software and operating systems aren’t written to run on the ARM architecture. But that’s changing as Ubuntu, KVM and others pledge to support ARM. This issue also may become less necessary in a world with software-defined data centers.

“The whole notion of software legacy as an obstacle diminishes more and more over time,” says Evans. “Your server isn’t a piece of hardware, it’s a service running software targeted to higher and higher levels of abstraction.”

In that model the server architecture becomes less relevant as layers of abstraction on top will knit the infrastructure together. And this seems to be happening, so now the only question for Calxeda and Evans is if he can duke it out in the developing microserver market. When I asked if anything had prepared him for the fight he’s taking on he replied, “Try developing a non-x86 architecture at Intel.”