Monday 11 August 2014

Mobile network energy in fridge units


I was boggled when I read in TIME: ‘The Surprisingly Large Energy Footprint of the Digital Economy’ [1] that an iPhone uses as much power as a refrigerator. I decided this deserves a deeper investigation. Obviously the iPhone itself is not using all this power – it is the mobile data streaming systems that download data to it.  However, this still seemed unlikely to me. It has taken a lot of digging to find the information I want but in the end I concluded it would need at least 30 iPhones to match the energy consumption of an American fridge. However, in the original source of the fridge comparison, a report from Digital Power Group (DPG) sponsored by the coal industry [2] the iPhone user was supposed to watch an hour of video each week. If you do that then the fridge comparison is not quite so ridiculous because video needs a lot of data. You only need about 4 iPhones to match the fridge.


By the way, Americans use much bigger fridges than we do and they use more power – TIME assumes 350 kWh/year. Henceforth I will base my units on 120 kWh/year which is reasonable for a UK A+ under the worktop fridge.

Estimates of energy for data transfer vary by a factor of 100
The refrigerator comparison in TIME is based on a heavy iPhone user downloading 1.6 GB/month and a network energy of 19.1 kWh/GB. At that rate, the total mobile energy use would be 367 kWh/year – more even than the American fridge. However, quite apart from being implausible, the 19.1 kWh/GB is an interpretation of a slightly different statement in The Mobile Economy 2013 [3]. The actual text reads ‘19.1 kW/GB’ but this does not make sense. It doesn’t have consistent units because kW is a rate and GB is a quantity. TIME has assumed this was just a typo for kWh/GB. It could also mean 19.1 kW/(GB/s) or 0.005 kWh/GB. That seems much too low.

DPG quotes an alternative estimate of 2 kWh/GB from a paper from the Centre for Energy Efficient Communications (CEET) [4]. I suspect it is also a mistake because CEET actually quotes 73 µJ/bit which works out as 0.2 kWh/GB – ten times smaller. But when I looked at the CEET paper I came to the conclusion that this figure is on the optimistic side.

Energy efficiency improvements of 26%/year – a leap of faith?
The CEET figure of 73 µJ/bit is for the power consumption of a 4G (LTE) mobile base station. This is quoted from yet another paper, this one from Energy Aware Radio and neTwork tecHnologies (EARTH) [5]. EARTH actually quotes 328 µJ/bit, for equipment which is commercially available in 2010. CEET reduces this to 73 by factoring in efficiency improvements of 26% each year to 2015 – an 80% drop over 5 years. It hasn’t pulled this factor out of the blue but it still seems to me to be quite a leap of faith. CEET estimates that in 2012 the equipment would be 24% better. That would turn 328 µJ/bit into 250 µJ/bit or 0.6 kWh/GB. I shall be conservative and stick with 0.6 kWh/GB.

Energy used depends mainly on the network capacity not the data usage
But when I looked in EARTH for the 328 µJ/bit I found this table.

TABLE 11. Aggregation to global scale for the considered traffic scenarios

Share of heavy usersECIP/AECIEBMean User Data Rate 5th perc. User Data Rate
#1: 20% heavy users0.654 kW/km20.615 kJ/Mbit46 Mbps14.4 Mbps
#2: 50% heavy users0.658 kW/km20.328 kJ/Mbit44.4 Mbps12.4 Mbps



The power per unit area (second column) hardly changes between the low and high use cases. This suggests power used depends on the area of coverage, not how much data is being transmitted. Actually it is the capacity that it is the key factor.

The figures in the table come from a particular scenario, with mobile base stations close together in urban areas where there are more users and further apart elsewhere. They have to be close enough so that they have the capacity to carry peak demand and this means most of the time they are running at low loads. Since most of the power consumption of each base station is fixed, regardless of how much data traffic it is handling, if you double the data transmitted you only add a little bit to the power consumption. At least, that is the way it works up to a point. If the base station does get overloaded, then you will find your data rate drops and if this happens too often you might complain to your provider. At some point they will decide they need to build some more base stations, and then the power consumption goes up.

In summary, the power needed to run the mobile network depends mainly on peak demand, not total demand. The 0.6 kWh/MB is the average for a lightly loaded base station and your mobile provider might decide to use fewer stations, and hence less energy, than this, at the risk of occasional overloading. So the 0.6 kWh/MB could be on the high side for two reasons – because the network has been optimized for a lower capacity or because the equipment is now more efficient. It is still a lot less than 19 kWh/MB. At this rate, a heavily used iPhone uses 11.5 kWh/year, so 10 iPhones to a fridge (UK fridge).

49 MW for mobile coverage of the UK - about 3.5 million fridges.
The total area of the UK is 244,000 km2 of which about 40% is significantly populated. Mobile coverage of this area at 0.5 kW/km2, (assuming 24% efficiency improvement on the 2010 figure from CEET) would require 49 MW. That means the mobile internet could use as much as 3.5 million fridges.

Your WiFi network could use nearly as much as your fridge
If you have a WiFi network in your house, that also consumes power all the time, regardless of whether you are actually streaming data on it it or not. The BT Infinity Hub (measured by a friend of mine) takes about 12W. This is quite a high end device and you can get ones that use a lot less than that. However the BT infinity uses nearly as much power as a fridge.

If you are worried about this, you could turn your WiFi hub off at night when you are asleep and/or get a less powerful hub.

Data transfer over the wired internet? – not very much
By the way, the wired internet also consumes power to route and transmit data. An interesting study based around video streaming for a conference co-located in Switzerland and Japan concluded that the energy used was 0.2 kWh/GB [6]. As for the mobile network, the power consumed by the internet relates to the capacity of the network rather than the traffic. In this study they portioned out the energy use by each component according to the fraction of the load at that time. Using this method it turned out that half the power was used by very lightly loaded equipment at each end of the route, because they used dedicated routers for the conference. If you exclude those components, then the total was only 0.1 kWh/GB. Also, this was for a 20,000 km route because they took the long way from Switzerland to Japan via the USA to avoid network congestion. The CEET study estimated that the core and metro internet uses much less than this: just 0.64 µJ/bit (0.0014 kWh/GB). Well let’s split the difference and call it 0.05kWh/GB, a twelfth the energy for the mobile internet.

Watching a film on the TV streamed over the internet once per week is 0.2 fridges. If you stream it 4G mobile to your tablet it would be 1.3 fridges.
Suppose we watch an HD film streamed over the internet (from Love Film or Netflix). It takes 2 hours and requires 4.6 GB data. The energy for the internet streaming is 0.23 kWh. Then you have to consider the two ends – the servers generating the data and the device you are watching it on.

A study of video streaming in the US [7] suggests the servers use just 0.0011 kWh/GB. (This is less than the CEET estimate for data centres in general but video streaming doesn't involve a lot of computation, just shovelling data from disk to network).

if you display this on a 100 W TV rather than a 3W tablet you use another 0.2 kWh so we must add that in.

Alternatively, you could be old fashioned and use a physical DVD. The energy needed to make a DVD including case and sleeve is estimated to be 0.46 kWh [7]. That is rather more than the energy needed to stream the data over the internet but if you watch the DVD several times then you can save energy this way.

It is all rather complicated so here is a table comparing four ways to watch a 2 hour film in units of fridge-weeks.



Ways to watch a 2 hours HD film (4.6 GB)Total energy (kWh)Total energy (fridge-weeks)Calculations
TV and physical DVD which you watch 3 times in all.
0.35
0.15

TV (100W) 0.2 kWh
DVD 0.46 kWh/3 = 0.15 kWh
TV streamed over the internet.
0.43
0.19

TV (100W) 0.2 kWh
Internet 4.6 GB at 0.05 kWh/GB = 0.23 kWh
Tablet streamed over the internet on your home WiFi network.
0.24
0.10

Tablet (3W) 0.006 kWh
Internet 4.6 GB at 0.05 kWh/GB = 0.23 kWh
Home WiFi – nothing because it doesn’t use extra energy for data
Tablet streamed over the mobile internet through your smart phone .
3.0
1.3

Tablet (3W) 0.006 kWh
Phone (2W) 0.004 kWh
Internet 4.6 GB at 0.05 kWh/GB = 0.23 kWh
Mobile 4.6 GB at 0.6 kWh/GB = 2.8 kWh

UK fridge for a week.
2.3
1
120 kWh/year
American fridge for a week.
6.7
3
350 kWh/year



Most of us don’t use the mobile network to watch films, yet. The ‘heavy use 1.6 GB/month’ comes to 0.2 kWh/week – just a tenth of a UK fridge. However, with 4G networks we can watch videos while travelling if we want - an HD 2-hour video once a week comes to rather more than a UK fridge but still a lot less than an American fridge.

If you are concerned about energy use for mobile, avoid using it for HD video, at least at peak times. If you want to save energy at home, check your Wifi hub and get a more efficient TV. You can find many 32” LED TVs these days using 60W or less.


[1] The Surprisingly Large Energy Footprint of the Digital Economy [UPDATE] (TIME) 14/August/2014


[2] Mark . Mills, The Cloud begins with Coal(Digital Power Group) August 2013

[3] A.T. Kearney The Mobile Economy 2013 (GSMA)

[4] The Power of Wireless Cloud (CEET) June/2013


[5] Energy efficiency analysis of the reference systems, areas of improvements and target breakdown (EARTH) Dec/2010


[6] Vlad C. Coroama, Lorenz M. Hilty, Ernst Heiri, and Frank M. Horn,
The Direct Energy Demand of Internet Data Flows (Journal of Industrial Ecology volume 17 number 5) 2013

[7] Arman Shehabi, Ben Walker and Eric Masanet The energy and greenhouse-gas implications of internet video streaming in the United States (Environmental Research Letters) 4/Apr/2014

2 comments:

  1. Really nice piece. It merits a (shorter) letter to Time, setting the story straight, and leading readers to this blog.

    ReplyDelete
  2. This comment has been removed by a blog administrator.

    ReplyDelete

Comments on this blog are moderated. Your comment will not appear until it has been reviewed.