Are internet data centres inefficient power hogs?

Internet data centres are increasingly being labelled as inefficient power hogs exacerbating global warming. If this is so, then should we encourage users to restrict their internet use? No. The solution is investment in more efficient data centres, and this requires the revenues generated by robust demand.

Global internet traffic is estimated to be growing at around 50–60% each year, with current users being the tip of the iceberg — comprising only a quarter of the world’s population. There is a lot more growth in user numbers and traffic volumes to come in the future, as new users come online in Africa, Asia, the Middle East and South America.

This growth scenario is starting to attract attention in terms of the energy inputs and outputs of the massive data centres that power the internet. The simple, and apparently benign, act of initiating an online search, updating a blog or chatting on a social network inevitably causes servers to heat up a little bit more in a data centre somewhere — and a power station to spew out more CO₂ to generate the electricity.

Researchers are now starting to attempt to quantify the energy impact of internet usage, with Google being fingered as an obvious target. A Harvard University researcher estimated that a Google search generates something like 7 grams of CO₂. Google engineers rebut this with an estimate of 0.2 grams.

The science of this measurement is ‘imprecise’ but let’s put it in perspective. A small car generates something like 100 g of CO₂ per kilometre, so a Google search is equivalent to driving a small car between 2 and 70 metres.

Many people avoid using cars to minimise their CO₂ footprint so, using the same logic, should we also now shun the internet to protect the planet?

The IT industry should look to increase supply-side energy efficiency. Making people aware of the energy impact of their daily activities is a useful thing. After all, no individual raindrop believes itself to be responsible for the flood, yet their cumulative effect can be disastrous.

However, stifling demand for the internet should not be seen as a solution to internet data centre energy consumption until we have made more progress on at least three areas of supply-side energy efficiency.

The first is power usage efficiency (PUE). The US Environmental Protection Agency’s 2007 study revealed that the historical PUE of data centres was observed to be around 2 — meaning that half of the energy that flows into a data centre is wasted in such things as power conversions, cooling and lighting before being applied to useful computing work. State-of-the-art data centre engineering and server consolidation/virtualisation strategies can see PUE reduce to 1.2 and below, but this requires investment in refurbishing legacy data centres and server farms and in new operational practices.

The second is the low energy proportionality of the current suite of computing technologies. Reducing demand by 10% does not necessarily lead to a 10% reduction in energy consumption. CPUs and RAM chips consume power even when idle. Disk drives spin regardless of data traffic. There is a lot of R&D underway to improve the energy proportionality of data centres, but the reality today is that reducing demand does not necessarily lead to lower CO₂ emissions.

The third is the often hyped ‘green energy’ — connecting data centres to solar, wind or geothermal power to replace coal-fired generation.

Here is the catch 22. Robust demand is required to generate revenues to fund investment in the more energy-efficient technologies required to actually reduce global data centre CO₂ emissions. Too much hype around the CO₂ cost of internet usage may well lead to people choosing to shun the internet, starving companies of the traffic-related revenues needed to actually solve the problem. Perhaps it’s better just to suggest that people turn their computers off when they are not being used.