James Hamilton, VP & Distinguished Engineer for Amazon Web Services, discovered the potential of Amazon in the world of cloud services when he realized how it could improve customers’ experiences and results, as he explained in an interview with theCUBE co-hosts John Furrier and Dave Vellante, live at AWS re:Invent 2013.

“I saw opportunity to improve by at least a factor of 10,” Hamilton said. After the S3 announcement, “I started running apps against it, saw it was super reliable. This is the beginnings of when I knew this would change the world.”

The evolving landscape of the data center

Commenting on the statement that in data center costs, the top expenses were servers, cooling power, and power itseff, Hamilton agreed that servers dominated, followed by power for cooling, and then actual power. Furthermore, the dominant cost of offering a service was that of the facility itself with “will run two million before there’s service in it. If you build a lot of data centers, you get good at it.” AWS data centers are very reliable and “costs are plummeting fast,” another added benefit for customers being that they “don’t need to predict years in advance what we need.”

Asked through CrowdChat about the impact of ARM in the data center, Hamilton said “I am super excited about ARM, it’s early days. I love seeing lots of players, I love seeing lots of innovation.” ARM isn’t just about every device everyone is carrying around, it’s broadly deployed. AWS is interested in ARM, “we are investigating deeply,” but, Hamilton stated, “at this point it hasn’t happened yet.”

Discussing how splitting workloads up changed utilization factor, Hamilton said that the price price performance of microservers got exciting in 2007 when “it was easy to produce a win. There are lots of parallel workloads where there is room for a big gain.”

As far as Spot instances’ role in helping create the world’s most cost effective super computer, Hamilton explained, “I am super excited by Spot,” it was one of the reasons he joined Amazon. “It’s taking the ideas from financial markets when you have high value assets. Poor utilization is the elephant in the room.” Only 15-20 percent or server capacity is generally used. 70 percent is wasted. Spot works to avoid such wastage. Hamilton agreed however that Spot is not recommended for every scenario. “Workloads that don’t operate well into an interrupted environment should not be run in Spot.”

Asked to comment on commodity networking and merchant silicon, Hamilton said the whole industry had needed network commoditization for 15 years. Cost is falling everywhere, except networking. “It’s really changing the Amazon business and what we can do for customers.”

Discussing specialized software and running general purpose processors, Hamilton explained that “today in AWS we run tens of thousands of copies of a single type of server. Now we’re pulling more down into the hardware and do more customization for our customers.”

“One of the problems when you’re building a server, is it could end up anywhere,” Hamilton said. The servers have to be designed with huge headroom on cooling requirements. “We take a different approach.” They work on the assumption that they go to good data centers and build servers that run in them.

Asked what the next big thing in the large scale paradigm of computing and computer science was, Hamilton said it was power distribution, with “tons of work going on in the field.” Networking was another area seeing a lot of action.

Discussing the issue of version control and completely updating the stack, Hamilton stated it was “super important form a security perspective that the code be updated and current.” Also from a customer point of view, this process has to be non-disruptive. “We monitor the heck out of our system,  our goal – we need to know any issue before the customer knows it.”

Asked to share his opinion on make vs buy, Hamilton said “it changes every few minutes. Years ago, when I joined the comp, we were buying servers from OEMs. Now we have our own, custom designed. Now server components are customized for us. Every year it’s a little more, it’s specialized.”

“I know exactly how much infrastructure we put everyday,” he added, the number is mind boggling. “Every day we deploy enough servers to support a 7 billion dollar company. My definition of scale changes every day.”

Asked to create a bumper sticker text for this year’s re: Invent event, Hamilton chose “It’s real, baby!”