What does it mean? What does Circle Economy (a nonprofit) suggest? Can and should we outsource the computing power of our personal devices to the cloud? It sounds climatically reasonable (it's more efficient, less direct and indirect GHGs, I take it) but I'm not sure it's not putting your eggs in few baskets
2 Answers
If you put aside the big issue of confidentiality of your activities on the outsourced device and of your data, it sounds good for several reasons :
It lets you use efficient devices managed by professionals : Frequent software updates and quick support.
Those devices would be used as much as they can be : Few computers in standby, lifetime as long as profitable, etc... That is less consumption of hardware.
However, any action you make (reading/writing in files, using large softwares, etc...) would need sending and receiving data from you personnal terminal (which could be a raspberry pi), which actually consumes energy.
I think a compromise can be found, but the balance between advantages & disadvantages is unclear to me.
- 131
- 2
-
2Another point you have to put aside is access to your own data: connections aren't perfect and even the most reliable service has downtime – Chris H Dec 21 '21 at 13:48
-
@ChrisH that's more of a budget question, even in remote areas it's possible to get reliable internet if you're willing to pay (Starlink is the cheap version of satellite internet, for example). We run two wired connections plus cellular at work, just so we're more likely to have connectivity if there are problems. At home wired+celleular is easy enough to set up, USB 4G dongle are common. – Móż Dec 22 '21 at 07:02
-
@Móż, to some extent that's true. I've got fibre to the cabinet which is pretty reliable apart from brief glitches; only my phone can fall back to 4G. But the cloud services themselves can and do fail (hopefully only temporary outages, but Google et al. also switch off useful APIs on a whim), and running a backup connection costs electricity as well as money. – Chris H Dec 22 '21 at 17:32
-
Yes, but then so does running RAID, the cost difference between a Raspberry Pi with a uSD card and something with enough power to run RAID and the whole rest of your failure-tolerant setup is significant. Both in power and money terms. I've run parallel Pi setups before and they both died to the same software flaw at the same time, leaving their uSD cards corrupted. Luckily they'd been posting data to another machine that didn't crash. – Móż Dec 23 '21 at 06:31
This totally depends on the amount of computation performed in the cloud versus the latency and data transfer amount.
For small quick computations, latency kills the idea of using the cloud: it's faster to compute locally than to transfer the data to a cloud that will most likely be at least 50 milliseconds round-trip away from you. So it only makes sense if the computation performed locally would take nearly a second or more.
For large computations, the key is to determine how much data transfer is needed to give the server in the cloud enough data to do the computation. Then you need to calculate how much carbon dioxide emissions you save from requiring a smaller battery and perhaps a smaller CPU in the mobile device that offloads its computations to the cloud versus how much energy the mobile network base stations use to transfer the needed data. For massive datasets, it may well be useful to perform the computation locally. However, for datasets where the problem description and answer description are short (for example please give me the fastest car route between these two points) but the computation would use lot of battery on the mobile device, it's likely to make sense to perform the computation on the cloud.
Don't underestimate the power use of the mobile network base stations!
- 5,010
- 6
- 17
-
Are cloud batteries more efficient than cell phone batteries (the Google Maps case you described)? – Sergey Zolotarev Dec 22 '21 at 00:44
-
@SergeyZolotarev most cloud systems run on grid electricity and only have batteries as backups in case the grid is unavailable. So their "battery efficiency" is nearly 100%. Like everything else, it goes up because the small cost is split across a lot of computing poiwer. – Móż Dec 22 '21 at 07:03
