Are Green IT experts addressing the wrong problem? Last month, a report said that in the public cloud, end user access over wireless networks uses ten times as much energy as the data centre. However the report, from Australia’s Centre for Energy Efficient Telecommunications (CEET), has come under strong attack from experts. Kate Craig-Wood, managing director of British hosting company Memset took issue with the numbers – and here she explains why:
As well as my day job running Memset, I’m a part-time PhD student researching cloud computing in relation to climate change, specifically the question of whether Moore’s Law (in its broad sense) will sufficiently mitigate the Rebound Effect in relation to ICT to allow ICT to fulfil its promise of greening society without itself becoming a major emitter. I am also chair of Intellect UK’s climate change group and a member of the ministerial Green Economy Council.
My PhD supervisor and I have just finished a body of work on energy cost of internet mediated transactions, including the network. In pursuit of that, we built a model to determine the energy per byte for three scenarios (general cloud services consumed at home, media streaming at home, and general cloud services consumed on the hoof). We are in the process of submitting the paper to a peer-reviewed journal so the work is not yet public but I can can share some elements. The Technology Strategy Board is also funding work in this area via the ESP KTN (the knowledge transfer network for electronics, sensors and photonics).
A key finding of our research was that the energy consumed in networks is utterly inconsequential to the energy consumed at the data centre and the energy consumed in the home (for general cloud services).
Further, for high-bandwidth activities like media streaming almost all the energy is consumed in the home (93 percent of it if you include the phone line ad smart TV). Of that the home wireless access point (WAP) is a significant proportion admittedly (about 25 percent of the total) but here common sense must apply: home broadband router-WAPs use less than 10 watts – at most 240Wh per day for a household. If that household likes hot drinks and there are 4 in the family, that’s about the amount of energy the kettle needs to make a round of tea each morning.
In further response to to the CEET paper regarding mobile phone networks, a quick sanity check helps. According to this learned IET article a UK 3G network uses about 12,000 base stations, and they use about 500Watts each. This is for one mobile network – the Mobile Operators Association (MOA) says there are 52,500 mobile phone masts in the UK.
500W * 52,500 base stations * 24 hours / 30m adults ≈ 21 Wh / British adult / day
To put this in perspective, the average Briton consumes 195 kWh/day of energy. This suggests that mobile base stations account for 0.01 percent (one hundredth of one percent) of a typical person’s energy budget – not exactly something to get excited about given how deeply integrated mobile phones are into our daily lives. It should be noted in passing that mobile cell basestations in some parts of the world are more carbon-intensive because they are not powered by mains electricity but rather by diesel.
Our own research suggests that the ratio of energy consumed by the data centre to the 3G basestation in a mobile cloud transaction is more like 10:1 than CEET’s 1:10.
However, it should be noted that the handset/phone itself dominates the energy-per-byte equation accounting for around 75 percent of the total. Of that 75 percent, however, 90 percent of the phone’s energy is actually its embedded energy, the energy put in to manufacture it (assuming a 2 year lifespan). Should we be demonising the phone, then? Over a two year lifetime a phone might use 150kWh of energy; which is less than one day of a British individual’s budget (0.1 percent of our personal footprint).
These figures come from Apple, a company which is more open about its energy than many commentators appear to believe.
So where did CEET go so wrong? I’ve been having a bit of a look in their appendix:
One clear issue is their assessment that a typical mobile phone uses 3Watts, which they appear to have scaled up and applied as the power requirements for a 4G connection. This is just plain wrong. If you take the battery out of your smart phone and read its label you will note that it says about 6Wh for a normal-sized one or perhaps 9Wh for a big one like a Samsung Note. If they used 3 Watts normal smartphones would discharge in 2 hours. They typically last more like 36 hours, so the power draw averages around 0.3 Watts (depending how much you use it).
Even if you take into account charging inefficiency (say 70 percent) that’s only about 0.4Watts – so CEET is out by close to an order of magnitude.
CEET also grabbed onto an ABI research figure suggesting that the number of wireless cloud service users would increase from 1.1 percent to 19 percent from 2008 to 2014. As part out our research we looked at the data usage of Memset staff – a small data set but still useful. On average our staff (all of whom are technically-minded and use cloud services on their smart phones) use 189MB/month (the median was 90MB/mo). If we then apply the 69 percent growth rate from the ABI research (which, by the way, I don’t think has been borne out over the last few years) to our own little sample group), it would suggest that by next year the average bandwidth consumed by mobile users would be about 320MB/month. The CEET paper assumes that by 2014 the average traffic per user will be 1.3-2.1GByte – a five-fold error.
They also assume that Wi-Fi power is a good measure of femtocells. First, the 2011 Cisco paper suggesting that 33 percent of handset and tablet traffic will be offloaded through “femtocells” seems wrong by common sense – as very few people are actually using them. What people are actually doing is using Wi-Fi hot spots a lot more. Admittedly that point is perhaps somewhat terminological but it does suggest a disconnection from reality.
However, as we have showed, if you use Wi-Fi hot spots and access points more, you decrease the total energy consumed per byte, you don’t increase it. CEET has taken the Wi-Fi energy and multiplied it up for the increase in traffic!
Home Wi-Fi cells are very inefficiently used; the average Brit consumes 23GB/month of broadband and if that’s all delivered over home Wi-Fi, using a WAP that consumes 10 Watts, it is roughly three orders of magnitude (1,000 times) less efficient than, say, a core Internet route. We have also shown that the power consumed by a home WAP is basically static (ie. it doesn’t go up if you use more data). If you put more data through that WAP and broadband connection it doesn’t cost you any more power since it is operating at a tiny fraction of capacity.
So greater use of femtocells and Wi-Fi means efficiency, not more energy.
Regardless, as described above, Wi-Fi hotspots are nothing to get in a flap about.
As for the data centres bit, yes, more power is going to data centres, but actually as an industry we are quite unusual, in that thus far we have demonstrated continuous and exponential improvements in energy efficiency. Similar rules apply to storage and networks; while we are consuming exponentially more, the efficiency of our storage systems and networks per-byte is also increasing exponentially. I have demonstrated this in my own early research, a relevant summary of which you can find here.
This is largely thanks to Moore’s law in its broadest application – the doubling of computer performance every 18 months per byte, and also per Watt. It is also thanks to the good works of experts (such as Romonet) in improving the mechanical and electrical efficiency of data centres. Further, we are progressively consolidating away from inefficient half-filled datacentres into modern Internet Data Centres (IDCs) the likes of which Amazon, Google, Facebook and my company Memset run. Given how vital they are to our day-to-day lives their energy consumption is pretty inconsequential. The ICT sector in Britain is also worth about 10% of GDP so we deliver good “bang for carbon-buck”.
When you factor in the fact that ICT is actually a major enabler of societal efficiency improvements, as described in various places, for instance Intellect UK’s high-tech low-carbon report, GeSI SMART 2020 “Enabling the Low Carbon Economy in the Information Age”, and WWF’s “saving the first billion tonnes”. Greenpeace also highlights the importance of “greening with ICT” on its Cool IT leaderboard.
It is a no-brainer; in exchange for our industry’s two or three percent of emissions, we have the potential to save 16 percent of emissions across society! But we do need data centres to do that.
It is illogical to suggest that all data centres should run off renewables. There is more demand for renewable-sourced electricity than there is supply, and there are plenty of other vastly more energy intensive industries that should be being persuaded to use green power. In Britain, for instance, nobody is investing in renewable power generation and that is a big problem for us.
Facebook is clearly the trailblazer for efficient data centre design with its Open Compute project and there is still much good work to be done there. I think it is perfectly reasonable for Facebook not to fret about sourcing renewable power while there is such a dearth of it. For Facebook energy is just a supply, not a core competence, and Facebook should instead focus on what it is expert in: efficient computing.
The issues generally come from people
Kate Craig-Wood is founder and managing director of Memset.
Are you a green IT guru? Take our quiz!