From Ewan Stern
James Clarage’s calculations of Google’s energy usage assume that all their “nearly a million” servers do is perform searches (3 April, p 20).
In fact, Google uses these data centres for many things: hosting YouTube and Gmail, caching just about every web page in existence and being one of the world’s largest web advertisers are just a few of them. Google is estimated to constitute between 10 and 30 per cent of the entire internet. Clarage compares the cost of a search to the cost of switching on a 100-watt light bulb. If I had to choose between the light bulb and Google’s many services, I’d happily live in the glow of a million LEDs.
From Tim Rustige
James Clarage states that “Google serves up approximately 10 million search results per hour, and so each search costs the same as running a 100-watt electric light bulb for an hour”.
Advertisement
According to his figures, then, Google serves up 240 million searches a day. In fact they serve up around 87.8 billion searches a month, around 3 billion a day.
It is my understanding that Google uses PCs for each server. I doubt that each of Google’s basic servers draws Clarage’s claimed 1 kilowatt of power; it would be something closer to 400 watts. And if Google has a million servers, as Clarage reports, then from his own figures, each one would only perform 10 searches an hour. This sounds like underuse to me. Maybe only a small fraction of those million servers search, while the rest serve up Gmail, Adwords or YouTube videos.
Knutsford, Cheshire, UK
James Clarage writes:
• The figures used in my estimate for the number of servers, and searches per hour, were for US traffic only. Worldwide traffic is of course higher, but so proportionally is the number of mirror-servers around the globe which handle this traffic. My kilowatt estimate for a server includes additional thermal factors, such as cooling, as well as the server’s power supply. Finally, the servers do not sit idle when not performing a search; they are busy downloading and indexing one of the world’s largest datasets.
London, UK
