Early this year, a newspaper story made the rounds when, after extrapolating from some energy use estimates made by an academic, it claimed that two searches on Google would burn enough energy to heat water for a cup of tea. The researcher behind the report eventually disavowed the newspaper's extrapolations but, by then, Google had already felt compelled to defend its energy efficiency in a blog post. Not content to let matters rest there, however, Google is back with a new set of figures, comparing searching using its servers to everything from a glass of orange juice to a cheeseburger. All of that may make the company feel better, but it misses the fundamental point: these sorts of figures are fundamentally misleading.
To be fair, just about anything with an environmental component these days announces some sort of comparison figure. The goal is fairly obvious: the public isn't accustomed to thinking in terms of kilowatt-hours or megajoules, so putting these numbers into a form that's more familiar will help efficiency figures resonate. These days, it's difficult to read about a large-scale project in, say, renewable energy generation that doesn't include figures on the number of houses it can power or the car-equivalents that are taken off the road.
And, for these large-scale projects, those numbers are pretty reasonable. You know how much power is coming out, and you know what an average home's electric use is, so it's easy to perform the calculation. Obviously, the numbers would be different if only Hummers or Priuses were being taken off the road, but, when the figure is in the tens of thousands, it's safe to expect that everyone recognizes it will involve a mix of typical car types.
Where things have the potential to go seriously awry, as we noted in our initial coverage of the issue, is when really small numbers and multiple rounds of estimation wind up getting used. Even a small error or a difference in significant figures has the opportunity to expand in magnitude as it works its way through the calculations. For Google's own numbers, which peg a search at about a kilojoule, we can't even examine the assumptions and errors, since they're simply based on what the engineers told the blogger (Urs H�lzle, Senior Vice President of Operations).
On the other side of the equation, one of the examples is a cheeseburger, which clocks in at 15,000 searches. That figure is apparently based on a lifecycle analysis of everything from the beef (which is very energy-intensive) to a few slices of pickle. Needless to say, the potential for errors to creep in at any step is enormous.
The other issue is that, when the comparison winds up being relative to a single item—a load of dishes, or a cheeseburger, among Google's examples—the tendency is to make that comparison with a specific instance of that item. For example, when it comes to the cheeseburger, are we talking gas, electric, or charcoal grills? It matters; a recent study found propane grilling to have a carbon footprint three times smaller than that of charcoal (again, on average). Meanwhile, Google's figures don't even bother with the average usage of a dishwasher—they take the maximum value allowed for the label. But here as well, specifics matter, as this number incorporates the average energy use for heating the water used during the wash, which will vary dramatically depending on how the hot water is supplied.
Unfortunately, Google has decided that these figures are so important, that they've now placed them on the pages devoted to describing how it achieves its energy efficiency. Most of the information in that section is actually useful, in that it provides specific information about how individual aspects of its server farms combine to make them very energy efficient. It's hard to see how fuzzy numbers about the searches per glass of orange juice help get that message across.
reader comments
26