4 ms·

The full report is here: https://groenlinks.nl/sites/groenlinks/files/2021-09/CE_Delft_210166_Carbon_footprint_unwanted_data-use_smartphones.pdf https://groenli

The full report is here: https://groenlinks.nl/sites/groenlinks/files/2021-09/CE_Delft_210166_Carbon_footprint_unwanted_data-use_smartphones.pdf https://groenlinks.nl/sites/groenlinks/files/2021-09/CE_Delf... I don't believe their numbers. They say that data use over wi-fi consumes 150 watt-hours per gigabyte transferred, or 220 watt-hours per gigabyte when including the energy used in data centers. Downloading Call of Duty: Black Ops Cold War, at 250 GB [1], would consume 37.5 to 55 kilowatt hours depending on whether you use the lower or upper figures. That's more energy than you need to fully charge a Nissan Leaf. Watching 8k video on YouTube can transfer up to 22.5 GB per hour [2]. At 220 watt-hours per GB transferred that implies 4,950 watts of power -- comparable to running a large air conditioner or an electric oven. If data transfer really had these huge power demands, you'd see electrical brownouts whenever a hot new streaming series debuts. When I've investigated reports like this before, the flaw of the methodology is usually something like: 1) Take a well-sourced megabits-per-watt study of networking equipment that somebody published long ago -- say, in 2003. 2) Measure today's data consumption. 3) Multiply the 2022 data consumption number by the 2003 energy intensity number. Get a huge result. Figuring out what happened in step 1 can be difficult because of indirect citations. A current paper will cite "Davy, 2018" but you look at the Davy references and it's getting numbers from "Smith, 2012" while Smith sources from "Chu, 2008" and Chu finally references the person who measured it: "Graham, 2003." [1] https://gamerant.com/pc-games-file-size-hd-space-biggest-huge/ https://gamerant.com/pc-games-file-size-hd-space-biggest-hug... [2] https://www.androidauthority.com/how-much-data-does-youtube-use-964560/ https://www.androidauthority.com/how-much-data-does-youtube-...

jonas21 4y ago
Yeah, many things in the report seems a bit off. Like they assume everyone uses the same amount of data regardless of what kind of network they're on. So since 5% of users are on 2G networks, they assume 5% of total data is transferred over 2G (which is ~400x less efficient per byte than 5G). > The EU average energy consumption of data traffic depends on the distribution of mobile networks used for cellular data traffic. The Ericsson mobility report also provides information on the type of smartphone subscription. This data is summarised in Table 3. We make the assumption that the distribution of the total data-use over the different mobile networks is the same. According to their numbers, 2G would account for nearly two thirds of total mobile energy use, which probably should've made them question their assumptions.
- p1necone 4y ago
  There seems to be a lot here that doesn't pass even a cursory sniff test. Just thinking about the power consumption of the networking equipment involved and doing some quick napkin math should have at least made them suspicious of their results.
  - philipkglass 4y ago
    You can spot-check a lot of claims with freshman level knowledge of the natural sciences, unit conversion, and arithmetic. But this combination of simple, powerful skills is less common than I would like.
BeefWellington 4y ago
> Downloading Call of Duty: Black Ops Cold War, at 250 GB [1] The rest of your comment is spot on but in the spirit of similar accuracy it's worth noting that that 250GB on the drive isn't how much is actually transferred. That said, while I'm sure energy improvements have happened and those numbers aren't correct I do think there's something to consider in the full equation of sheer number of pieces of equipment between you and the server. If we assumed 1wh per device per GB, that could sometimes be 10-20wh just for routers on your path. That doesn't include switches at the destination, server energy usage, etc. I have no idea what the number actually is on a per GB basis but just using that sample assumption as a way of illustrating that it could add up quite quickly. I suspect in reality it's nowhere near 1wh per device per GB transferred for routers, as even reasonably efficient consumer routers are 5-15W devices.
- p1necone 4y ago
  You also need to take into account the fact that most of this equipment is always going to be turned on and processing some non trivial level of load whether you're sending requests or not. There's generally going to be a difference in power consumption between different levels of load above idle state, but an n% increase in bandwidth is almost always going to result in < n% increase in power consumption. Also, any section of the network that's at full throughput all the time, and just evenly divides that throughput amongst users as it gets more load is not going to consume /any/ more power per GB no matter how many requests you throw at it.
  - toast0 4y ago
    > Also, any section of the network that's at full throughput all the time, and just evenly divides that throughput amongst users as it gets more load is not going to consume /any/ more power per GB no matter how many requests you throw at it. If it's at full throughput, you can't throw more load at it (well, smaller packets is more load than larger packets for the same throughput).
    - p1necone 4y ago
      I meant more at an individual level - like if I use 100gb of bandwidth today, and 200gb of bandwidth tomorrow there's a good chance that the extra power consumption on at least some of the network from my extra bandwidth is zero if what I described in my previous comment is the case. If everyone attempted to double their bandwidth usage tomorrow, then yes equipment is going to have to be upgraded. (Although even that could be a case of upgrading equipment that provided N throughput and cost Y watt hours before to newer equipment that provides 2N throughput, and still costs Y watt hours)
brigade 4y ago
Out of curiosity, I followed the wifi citation [1], and at a first reference it comes from the 1byte model on the page, which gives a simple Wh/byte figure. Which in turn appears be derived from the 5th spreadsheet by taking the total yearly energy consumption of devices that use wifi and dividing it by the amount of data said devices use over the year. Which I have absolutely zero idea what use that ratio even could have, let alone be important enough for an entire spreadsheet. But it certainly is complete garbage for calculating the energy cost of data transfers. (well, infrastructure kWh/GB calculated that way could be interesting, but still utterly useless for calculating a marginal energy cost) [1] https://theshiftproject.org/wp-content/uploads/2019/10/Lean-ICT-Materials-Liens-à-télécharger-réparé-le-29-10-2019.pdf https://theshiftproject.org/wp-content/uploads/2019/10/Lean-...

jonas21 4y ago

Yeah, many things in the report seems a bit off. Like they assume everyone uses the same amount of data regardless of what kind of network they're on. So since 5% of users are on 2G networks, they assume 5% of total data is transferred over 2G (which is ~400x less efficient per byte than 5G). > The EU average energy consumption of data traffic depends on the distribution of mobile networks used for cellular data traffic. The Ericsson mobility report also provides information on the type of smartphone subscription. This data is summarised in Table 3. We make the assumption that the distribution of the total data-use over the different mobile networks is the same. According to their numbers, 2G would account for nearly two thirds of total mobile energy use, which probably should've made them question their assumptions.

p1necone 4y ago
There seems to be a lot here that doesn't pass even a cursory sniff test. Just thinking about the power consumption of the networking equipment involved and doing some quick napkin math should have at least made them suspicious of their results.
- philipkglass 4y ago
  You can spot-check a lot of claims with freshman level knowledge of the natural sciences, unit conversion, and arithmetic. But this combination of simple, powerful skills is less common than I would like.

BeefWellington 4y ago

> Downloading Call of Duty: Black Ops Cold War, at 250 GB [1] The rest of your comment is spot on but in the spirit of similar accuracy it's worth noting that that 250GB on the drive isn't how much is actually transferred. That said, while I'm sure energy improvements have happened and those numbers aren't correct I do think there's something to consider in the full equation of sheer number of pieces of equipment between you and the server. If we assumed 1wh per device per GB, that could sometimes be 10-20wh just for routers on your path. That doesn't include switches at the destination, server energy usage, etc. I have no idea what the number actually is on a per GB basis but just using that sample assumption as a way of illustrating that it could add up quite quickly. I suspect in reality it's nowhere near 1wh per device per GB transferred for routers, as even reasonably efficient consumer routers are 5-15W devices.

p1necone 4y ago
You also need to take into account the fact that most of this equipment is always going to be turned on and processing some non trivial level of load whether you're sending requests or not. There's generally going to be a difference in power consumption between different levels of load above idle state, but an n% increase in bandwidth is almost always going to result in < n% increase in power consumption. Also, any section of the network that's at full throughput all the time, and just evenly divides that throughput amongst users as it gets more load is not going to consume /any/ more power per GB no matter how many requests you throw at it.
- toast0 4y ago
  > Also, any section of the network that's at full throughput all the time, and just evenly divides that throughput amongst users as it gets more load is not going to consume /any/ more power per GB no matter how many requests you throw at it. If it's at full throughput, you can't throw more load at it (well, smaller packets is more load than larger packets for the same throughput).
  - p1necone 4y ago
    I meant more at an individual level - like if I use 100gb of bandwidth today, and 200gb of bandwidth tomorrow there's a good chance that the extra power consumption on at least some of the network from my extra bandwidth is zero if what I described in my previous comment is the case. If everyone attempted to double their bandwidth usage tomorrow, then yes equipment is going to have to be upgraded. (Although even that could be a case of upgrading equipment that provided N throughput and cost Y watt hours before to newer equipment that provides 2N throughput, and still costs Y watt hours)

brigade 4y ago

Out of curiosity, I followed the wifi citation [1], and at a first reference it comes from the 1byte model on the page, which gives a simple Wh/byte figure. Which in turn appears be derived from the 5th spreadsheet by taking the total yearly energy consumption of devices that use wifi and dividing it by the amount of data said devices use over the year. Which I have absolutely zero idea what use that ratio even could have, let alone be important enough for an entire spreadsheet. But it certainly is complete garbage for calculating the energy cost of data transfers. (well, infrastructure kWh/GB calculated that way could be interesting, but still utterly useless for calculating a marginal energy cost) [1] https://theshiftproject.org/wp-content/uploads/2019/10/Lean-ICT-Materials-Liens-à-télécharger-réparé-le-29-10-2019.pdf https://theshiftproject.org/wp-content/uploads/2019/10/Lean-...