While the image that the high-tech industry has adopted for the source of services delivered over a remote network — the cloud — conjures thoughts of airy skies where quasi-magical things happen, the actual cloud is firmly anchored to the ground in data centers bristling with computer servers. As the cloud continues to expand, so does the demand for servers — servers that consume more and more resources like electricity, space and capital.
What’s to be done? An obvious answer is to create servers that consume less power, take up less space and cost less to deploy. And that’s what a company called Calxeda says its new EnergyCore “server on a chip” will do for servers that use it.
Calxeda announced its EnergyCore processor Tuesday at simultaneous events held in Austin, Texas, where the company is headquartered, and Palo Alto, Calif., where HP revealed its intention to include the chip in its first generation of extreme low-energy server development platforms.
“Most large enterprises, and even Web-scale companies, have an insatiable need for more power in the data center,” Calxeda Director of Product Marketing John Mao told TechNewsWorld, “and the top constraining factors that they face to obtain that power is the lack of energy running into the buildings and the amount of space that they have.
“Data center operators will tell you that they have plenty of money and budget to buy more servers, but they can’t get them physically deployed because of those two reasons,” he added.
Bulldozers and Shovels
To address those problems, what companies like Calxeda have done is look to a sector where optimization of power consumption and space is a must: the cellphone industry. That’s why Calxeda has based its EnergyCore processor on the ARM technology used today in many cellphone processors.
While chips based on the ARM architecture don’t have the muscle packed into the x86 processors in a typical server, for many of the functions taking place in the cloud, such power is overkill.
“We take an approach that says, ‘You don’t need to buy a bulldozer when all you need is a shovel,'” Mao explained.
The result is EnergyCore, which can be used to build servers that consume 90 percent less energy and take up a fraction of the space of their conventional counterparts. In a design unveiled at the HP event, for instance, the computer maker showed how 2,800 servers based on Calxeda’s chips could fit in the space of 40 of today’s servers — and at the same time consume less energy than those servers.
“That means that all of a sudden, these data center operators have increased the performance in their data centers 10 times without having to expand their facilities and while cutting their energy consumption,” Mao said.
32-bit Versus 64-bit
Calxeda isn’t the only company making processors for low-power consumption, high-density servers. SeaMicro makes such a chip based on another mobile processor, Intel’s Atom chip. Unlike Calxeda, which won’t start shipping its EnergyCore product until the second half of next year, SeaMicro already has servers based on its processor on the market.
A problem with a server based on ARM technology is that it doesn’t support 64-bit processing, according to SeaMicro CEO Andrew Feldman. Because ARM only supports 32-bit processing, it can’t address as much memory as a 64-bit processor, he explained.
“Each core gets only one gigabyte of memory,” he told TechNewsWorld. “That’s very little memory.”
Calxeda’s Mao, however, discounted the need for a 64-bit processor. The reason nearly all data centers use servers with 64-bit processor is that they can’t get their hands on servers with 32-bit processors, he opined. “There have been a lot of end-users that have come back to us and said, ‘You know what? Actually, 32-bit will work.'”
However, “it won’t work for everything,” he conceded. “But where it matters for efficiency, 32 bits is just fine. In fact, it may be better because it’s more efficient than what they’ve seen on their other platforms.”
Peter ffoulkes, research director for servers and virtualizaton for the 451 Group in Boston, also frowned on the importance of 64-bit support.
“It’s not that big of an obstacle,” he told TechNewsWorld. “There’s an awful lot of people still running 32-bit applications.”
Moreover, the kind of jobs the processors would be called on to do could easily be distributed across a number of 32-bit processors just as well as a number of 64-bit ones, he explained.
“You don’t need a big 64-bit memory space to do that,” he said. “You just give each processor two gigs of data or four gigs of data at a time and let them work on it.”
At the HP event, the company announced an initiative called “Project Moonshot” designed to fuel the advancement of low-energy server technology while promoting industry collaboration to break new ground in “hyperscale” computing environments such as cloud services and on-demand computing.
For companies with thousands of servers delivering Web services, social media and simple content delivery applications, HP said, Project Moonshot is designed to deliver improved simplicity while achieving energy and cost savings never before possible.
“We want to advance computing, but we want to advance computing at a massive scale and deliver quantum leaps in the efficiencies that people have today,” Jim Ganthier, HP vice president for marketing and operations for the Industry Standards and Service Group, told TechNewsWorld.