To an internet based company, server infrastructure is the secret sauce that can really help a company pull ahead of its competitors. It determines the quality of service its customers will receive, and their cost and efficiency will have a huge impact on the bottom line of the company. To Google this secret was a carefully guarded one, with few outside of the company having any real details. The only thing that we knew for sure is that they were built in house using parts that are generally available to every PC builder. This all changed last week when Google decided to lift the veil of secrecy to a group of IT professionals.
Each server measures about 3.5 inches thick, and is designed in a custom rack for easy stacking. Each unit sports two x86 processors either from Intel or AMD, contains two hard drives (presumably configured as a raid 1), and eight memory slots. These components are mounted on a Gigabyte motherboard, and protected by built in 12-volt battery that also serves as a UPS.
The built in battery was perhaps the biggest secret that was revealed and is a slap in the face to traditional thinking when it comes to large scale battery backup. Typically, server farms employ massive uninterruptable power supplies in the event of a power failure. The biggest problem with this approach according to Chris Malone from Google is the ability to scale it perfectly for the number of servers, and inefficiencies inherent with the technology. “Large UPSs can reach 92 to 95 percent efficiency, meaning that a large amount of power is squandered. The server-mounted batteries do better, Jai said: "We were able to measure our actual usage to greater than 99.9 percent efficiency."
Google’s approach to server infrastructure is defiantly unique, and it’s use of low cost customer grade hardware defiantly helped them survive the early years on razor sharp margins.