At GoogleGOOG +0.53%, hundreds of fans funnel hot air from the computer servers into a cooling unit to be recirculated at a Google data center in Mayes County, Okla.
Associated Press
A few months ago, James Weatherell updated his online resume on LinkedIn with his recent experience working with new cloud-computing software known as Open Stack.  Since then, the phone of the 27-year-old computer engineer who works at the Broad Institute of Harvard and the Massachusetts Institute of Technology has been ringing off the hook.
“I get three or four calls a day from headhunters,” said Weatherell.
Weatherell’s mojo reflects the growing popularity of Open Stack, a new type of open software that more companies are using to manage their private cloud-computing systems. In December, Weatherell helped the research institute to create a proof of concept using Open Stack software and computer servers from Cisco SystemsCSCO +1.41%.
The new system, said Weatherell, is cheaper and more reliable than public cloud computing services from Amazon.comAMZN -0.05%, and also enables the institute to more easily protect its data and quickly add capacity to support the large computing needs of its researchers.
“It worked really well,” he said. “It made everyone feel this could work.”
The rise of so-called hyper-scale or high-performance computing in which companies increasingly manage thousands to hundreds of thousands of computer servers was pioneered by Google and FacebookFB -1.57%, but now most large companies and organizations are following their direction. Faced with exploding data growth and demand for computing gear to support it, Google and Facebook now build much of their own hardware, including servers that cost $1,500 apiece instead of the standard $5,000. Last year, Microsoft saMSFT +2.39%id the company had more than 1 million servers in its data centers.
It’s a shift that carries profound implications for the way technology is built and bought, the way software is developed and the future of the technology job market.
“Google figured out how to solve big problems using cheap computers,” said Mark Shuttleworth, founder of cloud computing company Canonical. “It sucked the importance out of paying for physical hardware or software that failed.”
Computer engineers like Shuttleworth and Weatherell often turn to an analogy to describe this shift. In the old style of computing, engineers say companies treated their computers like pets.  They spent a lot of money buying those servers, from thousands of dollars for low-end machines, to tens of thousands of dollars for proprietary UNIX servers or hundreds of thousands for mainframes.
Every server was given a name. And given their high price, companies spent large sums of money caring and feeding these precious machines. Armies of system administrators were hired to work day and night to make sure they served up Web pages for their users. If a server broke down, companies dispatched the administrators to fix the problem and try and bring the server back to life.
“In the old world you would go in, take the top off, and test the expensive memory,” said Weatherell, a former IBMIBM +0.33% systems administrator. “You needed to be an A+ information technology guy.”
But in the new world of hyper-scale computing, engineers treat servers like cattle. It doesn’t matter if a cow dies as long as the herd survives. Servers can be treated like throwaway items because they are much cheaper than they used to be, and because companies now have software that can connect these machines and enable engineers to shift tasks automatically from one box to another without a hiccup.
“Now you shoot servers in the head and leave them in the field,” said Joshua McKenty, a former technical architect of NASA who co-founded a cloud software company based on his work there called Piston Cloud Computing. “Eventually enough die and then you swap out the whole rack.”
Mr. Weatherell said his team “does not even bother supporting our hardware now.” His Open Stack system started out with about 80 cores, or computer servers, and is growing to 1,000 to 2,000 servers in the near term. Thanks to the new software, one administrator can now manage several thousand servers, instead of just 100 in the old model, say engineers.
The rise of the new massive computer networks is extracting the value out of computing hardware and making the software and expertise that runs these machines more valuable.
Tens of thousands of computer administrators and software developers will have to retrain themselves in the new arts of cloud software or risk obsolescence. But it’s also an opportunity for engineers that leap to the new world.
“Companies that jumped into cloud early are the kinds of companies” that headhunters are targeting, said Weatherell.