Google doesn't just know search; the company also appears to have a firm grasp on sustainability in its datacenter.
Take, for example, the average PUE (Power Usage Effectiveness) score for the search giant's datacenters. PUE compares the overall amount of energy used in a datacenter for all functions -- including computing, cooling, and power distribution -- to the amount that just goes into computing. (For PUE, lower is better, and 1 is as good as you can get.) Google reports an average weighted PUE score of 1.21 for all six of its datacenters.
That score means that just 21 percent of the power Google uses in its datacenters goes toward overhead such as cooling and electrical losses; the remaining 79 percent is put to actual work by storage, networking, and server gear. By the EPA's account, 1.2 represents a score for a state-of-the-art datacenter. Google is clearly doing a thing or two right.
But how, you might wonder, does Google do it? Fortunately, the company is willing to reveal at least some of its datacenter best practices in a recently opened section of its corporate site called "Commitment to Sustainable Computing." Here are some techniques offered by Google for wringing efficiency out of its datacenters:
1. Measure, measure, measure. If you wonder how your own datacenter stands up to Google in terms of efficiency, you should be calculating your own facility's PUE score on a regular basis. After all, how else can you possibly know how well you're doing? Gathering the data can require a little footwork as you walk among the machines taking measurements every so often (unless you have a sophisticated monitoring system in place like, say, Microsoft) -- but it helps illustrate which practices are working and which aren't.
[ There are caveats when it comes to measuring and interpreting PUE. Learn more by reading "Don't believe the PUE hype." ]
2. Use efficient machines. Google says it rates "the efficiency of our servers by measuring the power used by each of the actual computing elements (such as processors and memory) against the power used by all other things (like fans and power conversion)."
Sounds rather like a PUE score for hardware, doesn't it? And it makes abundant sense: How else could you know how efficient your hardware really is if you don't measure?