Some interesting theories and discussions have been bandied about the cybersphere recently, regarding the best way to make the web greener; for instance, many industry commentators have put forward the idea that moving 10Mb of data across uses the equivalent of 900g of coal in energy, and that changing the Google homepage to black would save massive amounts of electricity across the world. This idea has been lavished all over the web, most notably on Digg.
As a result of the rising concern over the "green web", many companies are becoming increasingly concerned with environmental damage and their energy costs. This move isn't entirely altruistic; after all, when memory and processing power are almost free and companies own massive server farms, electricity is the main cost of many large search companies to keep all their boxes running.

So when it comes to companies who deal exclusively in information and data, discussions will inevitably turn to the big daddy of them all - Google. As a company who are recognised for their liberal ethics, Google have always been very concerned about environmental problems, trumpeting their power-saving efforts and their use of ultra-low chips for all to hear. So we must ask exactly how much power would Google use up through the use of standard chips - and how much are they saving by switching to low power ones?
In order to establish this, first of all we need to guess how many computers Google use overall - strangely or not, this isn't public information. The best way to guess this is to estimate Google's capacity load by figuring out their market share and how much data they need to serve based on that.
Firstly - there is a population of 300 million in the US, and 60 million in the UK. We will assume that there is 70% internet usage amongst these people, and a 55% Google market share (this is a generous weighted average of the recently reported 47.4% market share in the US, and the 75% reported recently in the UK).
We will assume each user searches 10 times a day, and each SERP result page is 25k in size. Studies show that each user only looks at the first or possibly second page of results after making a search - making an average of 1.5 pages served per user. That's 375k served by Google per person per day.
So we can calculate that Google serves 360,000,000 (US and UK population) x 0.7 (user percentage) x 0.55 (Google market share aggregate) x 375k (data per person per day) - which is in totality 51,975,000 Mb per day.
The internet works by serving bits, not bytes, so by some conversion we see that Google serves 425,779,200,000,000 bits per day, which is 4.9 Gigabytes per second. That's quite hefty bandwidth.
The internet is built on a system that relies and expects data loss and failure - without getting too technical this means a single PC cannot serve 1 GBps (gigabytes per second) with a 1 GBps network card installed. In fact, the figure is much more likely to be 1MBps (megabytes per second) - so for 4.9 GBps we'd guess that there's approx 5000 PCs used to serve search queries alone. Now, this is for normal searches only. With image searches, we'd guess that with Froogle, Google Analytics, Google Earth, and the vast myriad of other searches alone, a conservative estimate would add another 50% of servers on to this figure. At a guess that's 7,500 servers for web based incoming queries alone, with no consideration of storage or Google crawlers.
Now we get to the real guts of the operation: the massive server farms needed to catalogue and index the internet. At this point, we have to be honest - it's quite hard to guess how many servers there are used by Google for this purpose, which makes an estimated guess necessary here. After some heady discussion around the bigmouthmedia office - involving C-blocks and D-blocks, IP addresses and covert missions into Google London dressed in black and white stripy jumpers - we came to the conclusion that there are probably around one hundred data centres in the UK and US. Some are much larger than others - most likely those located in London, Mountain View and Dublin - but that the average amount of PCs (taken as a mean) is around 700-750. These might seem like nice round numbers but a lot of thought went into them. That's about 72,500 servers. Of course, there is also the need to factor in the redundancy required to cope for dead drives and back-up storage - perhaps another 20,000 servers.
In addition, as of Dec 31, 2006, Google had 10,674 employees. They'll all have one PC each, plus internal network boxes. At a guess, that's another 12,000 PCs.
So, stretching the boundaries of the 'ballpark figure' convention, it seems to us that Google has about 12,000 personal sized PCs, and about 110,000 data servers of one type or another (this is a figure we've seen bandied about before, but we needed to establish it ourselves).
This is where we come to the crux of it. Now, say an average Google employee's box uses 200 Watts of power, and an average server uses 300 watts of power (factoring in surrounding issues like air conditioning), then over 24 hours you have 849,600,000 Watts - or 849.6 Megawatt hours per day. That's 310,104 Megawatt hours per year. By anyone's guess, that's a lot of electricity.
But, as a company, Google are extremely green. Google have started using very low wattage, extremely efficient chips; these suck nowhere near as much power as the massive, dual core chips that most web servers use, which are in the region of circa 130-150 watts. Generally, Google have been reported as using the lowest power chips they can find - generally the newest ultra-low power Opteron server chips, which use more in the region of 35 watts. As a compromise, as well as the fact that we know Google have a smattering of high power Intel chips still knocking around in their servers, we estimate a 40W average.
We can now say that, whilst Google can't realistically reduce the power in their internal PCs, they certainly can with their servers, and with low voltage solid state hard drives, totally bare rack-frame minimal architecture and low power CPUs, their servers probably only use more along the lines of 150 Watts.
Factoring that in, we can say that Google probably uses around 424.8 Megawatt hours per day. That's a difference of 155,052 Megawatt hours per year. This amount is approximately 1/6th of the output of a large nuclear power station running at full capacity for a full calendar month. Of course, this is the power used by their computers only; there are also the multiple other energy costs that any large business must consider.
Yahoo! and MSN aren't quite so concerned with their green credentials. We guess that Yahoo! uses 49882.5 Megawatt hours per year, but could be using 23904.5 if it changed to low power chips - a difference of 25978 megawatt hours per year. What's more, MSN search (note - not the entire Microsoft corporation!) are in a similar situation, using 12455.6 megawatt hours; if they changed to low power chip architecture, they could be using 6,479.6 megawatt hours per year. This represents a difference of 5,706 megawatt hours a year - enough energy to power an entire town.
Google knows this. After all, they didn't become the world's leading search engine by being dumb. They've recently announced that they're covering every spare sky-facing surface in their Mountain View headquarters, the Googleplex, with solar power cells. Estimates say that this will supply 30% of the power needs of the complex - and all with free, completely environmentally friendly energy. Their data centres across the world are generally situated near large, environmentally friendly energy sources - for example, the Google data centre in Oregon sits right next to a hydroelectric dam, and buys large quantities of power at bulk rate directly from the dam.
Ultimately, the most poignant fact to consider is that a large Google data centre can use as much electrical power per day as an industrial aluminium smelter, with its hellish furnaces and 1000° Celsius temperatures. The major difference with Google is that they can change this - and that they care where their power comes from.


















