How cloud computing impacts the environment(thecloud.christmas)
thecloud.christmas
How cloud computing impacts the environment
https://thecloud.christmas/2019/15
66 comments
If you run your workloads in AWS us-west-2 or GCP us-west1 you're running on hydroelectric energy produced at the Columbia River.
These regions are also really inexpensive due to this fact.
These regions are also really inexpensive due to this fact.
Interesting thanks! Do you know about other locations like us-central1 for instance?
Hetzner uses hydropower. All Iceland hosts use geothermal.
https://www.hetzner.com/unternehmen/umweltschutz/
https://www.hetzner.com/unternehmen/umweltschutz/
I don't mean to nitpick, but wouldn't "as carbon neutral as it gets" be zero reliance on fossil fuels, even for backup?
Trying to use a carbon neutral energy source for emergency power is very likely to have a higher carbon footprint overall due to the fixed cost of a rarely used infrastructure.
For example, on a much smaller scale, before leds were so incredibly cheap, incandescent bulbs were more energy efficient overall than alternatives such as fluorescent lights for places that were rarely visited. Think cellars, attics, service closets ...
For example, on a much smaller scale, before leds were so incredibly cheap, incandescent bulbs were more energy efficient overall than alternatives such as fluorescent lights for places that were rarely visited. Think cellars, attics, service closets ...
Perhaps that’s not currently feasible, and thus you can’t ”get there” yet?
>as it gets
It's "as it gets" that's doing the heavy lifting here.
It's "as it gets" that's doing the heavy lifting here.
Agreed. And "it" is the real workhorse - is "it" carbon neutrality or datacenter power supply?
Zero reliance isn't currently feasible unless you're ok with constant outages. If you're connected to the grid you're using electricity from fossil fuel at some point.
Isn’t nuclear a fossil fuel as well? While not quite the same thing, I don’t think we can manufacture uranium (or whatever the reactor uses to run), more than once?
In the limit, energy is not a renewable resource, thanks to entropy. In that sense all energy is a "fossil fuel" left over from the Big Bang. However any Kardeshev type I civilization (located on one planet) will be satisfied with nuclear fission's power output for a time frame beyond our ability to predict into the future; that is, if we here in 2019 try to arrogantly predict that "there's a power crisis coming in 500 years!", it is very unlikely that 2519 is going to hail us as forward-thinking smart people, because the odds of something critical changing between now and then approach 100% [1]. Let alone several thousand years from now.
These predictions also generally include a healthy amount of assuming growth of energy in the future, too, not just projecting today's use indefinitely (although not necessarily predicting exponential growth either).
This also ignores what I would consider fairly good odds of non-terrestrial sources we may find on that time frame. Plus I'm still holding out hope that one of our fusion approaches will eventually pay off. (Dyson spheres are nice & all but there's still something to be said for being able to generate power in concentrated manners where you want it, rather than having to collect it at scale from an uncontrolled reactor.)
[1] For the fashionably misanthropic among us who may feel that statement may contain too much optimism, "total civilization collapse" would be such a "critical change".
These predictions also generally include a healthy amount of assuming growth of energy in the future, too, not just projecting today's use indefinitely (although not necessarily predicting exponential growth either).
This also ignores what I would consider fairly good odds of non-terrestrial sources we may find on that time frame. Plus I'm still holding out hope that one of our fusion approaches will eventually pay off. (Dyson spheres are nice & all but there's still something to be said for being able to generate power in concentrated manners where you want it, rather than having to collect it at scale from an uncontrolled reactor.)
[1] For the fashionably misanthropic among us who may feel that statement may contain too much optimism, "total civilization collapse" would be such a "critical change".
Thanks, that (together with other comments) was actually really informative.
"Fossil" means "remains of plants or animals", but in the context of environmental aspect of fuels the key aspect is emission of CO2. Nuclear is neither of that. On the other hand it is non-renewable like fossil fuels.
In the longer term (hundreds of years), humanity should be worried exhausting non-renewable resources, but in a shorter perspective the climate change is a bigger concern.
In the longer term (hundreds of years), humanity should be worried exhausting non-renewable resources, but in a shorter perspective the climate change is a bigger concern.
As of 2009, estimates were that we have between 200-400 years of nuclear fuel remaining at current consumption rates.
If we can efficiently extract uranium from saltwater, that value jumps to 60,000 years.
If instead we switch all reactors over to breeder types instead of the current light-water reactors, our 200-400 year estimate then becomes more like 30,000 years
If we can efficiently extract uranium from saltwater, that value jumps to 60,000 years.
If instead we switch all reactors over to breeder types instead of the current light-water reactors, our 200-400 year estimate then becomes more like 30,000 years
Replace Uranium with Thorium and you get another increase in high-energy lifespan for humans by tens of thousands.
While yes, datacenters will be a significant consumer of energy in the foreseeable future, I honestly don't feel bringing "cloud" into the picture gives relevance other than the buzzwords.
Like what the article actually says, cloud is basically datacenter hosted by another guy. In fact, I would argue that cloud is more energy efficient due to the shared infrastructure and higher efficiency due to the scale involved. Imagine the amount of redundancy an organisation have to have to handle peak demand, vs a cloud provider aggregating the demands together.
The graph is probably even more telling. With (assumed) increased need for technology going forward, we actually have almost static demand for energy due to these technologies. It's actually a good news!
End of the day, yes technology is expected to consume more energy in the future, but in a good way, they will be somewhat aggregated to cloud. That means we can look into solutions targeted to cloud providers that actually have significant impact to the environment (eg. more incentives to use solar panels in cloud datacenter)
Like what the article actually says, cloud is basically datacenter hosted by another guy. In fact, I would argue that cloud is more energy efficient due to the shared infrastructure and higher efficiency due to the scale involved. Imagine the amount of redundancy an organisation have to have to handle peak demand, vs a cloud provider aggregating the demands together.
The graph is probably even more telling. With (assumed) increased need for technology going forward, we actually have almost static demand for energy due to these technologies. It's actually a good news!
End of the day, yes technology is expected to consume more energy in the future, but in a good way, they will be somewhat aggregated to cloud. That means we can look into solutions targeted to cloud providers that actually have significant impact to the environment (eg. more incentives to use solar panels in cloud datacenter)
> cloud is basically datacenter hosted by another guy. In fact, I would argue that cloud is more energy efficient due to the shared infrastructure and higher efficiency due to the scale involved
While I agree with this, I think it is precisely because of this commoditisation of cloud computing that we can have this conversation at all, as we have only 3 (AWS, GCP, Azure) companies to look towards to give a very accurate and measurable environmental impact of each CPU/TPU cycle or GB stored/transferred.
Prior to this "cloud" movement, I'd argue it was too fragmented to measure, other than as a cost of electricity.
My takeaway is that there are negative externalities to using computers beyond the direct costs, and that it is a good time to take stock of these and consider how to reduce them.
Edits:
From the article:
> However, there lies a responsibility on all of us developers as well to utilize libraries, coding techniques and compression algorithms which consume less storage and less energy. Mobile developers are well aware of the power restrictions due to batteries. It's time the rest of us follows and give their contribution to lowering data storage and processing for our workloads. The benefit? Reducing storage, memory and CPU for your workloads has an economic benefit - you pay less money to the cloud vendor. Win-win.
The issue will be in pushing an agenda that conflicts with the cloud providers motives (revenue & profit).
While I agree with this, I think it is precisely because of this commoditisation of cloud computing that we can have this conversation at all, as we have only 3 (AWS, GCP, Azure) companies to look towards to give a very accurate and measurable environmental impact of each CPU/TPU cycle or GB stored/transferred.
Prior to this "cloud" movement, I'd argue it was too fragmented to measure, other than as a cost of electricity.
My takeaway is that there are negative externalities to using computers beyond the direct costs, and that it is a good time to take stock of these and consider how to reduce them.
Edits:
From the article:
> However, there lies a responsibility on all of us developers as well to utilize libraries, coding techniques and compression algorithms which consume less storage and less energy. Mobile developers are well aware of the power restrictions due to batteries. It's time the rest of us follows and give their contribution to lowering data storage and processing for our workloads. The benefit? Reducing storage, memory and CPU for your workloads has an economic benefit - you pay less money to the cloud vendor. Win-win.
The issue will be in pushing an agenda that conflicts with the cloud providers motives (revenue & profit).
Disclaimer: I'm the author of the article
I don't really believe the suggestions conflict with cloud provider motives. If all cloud customers reduce their workload requirements, it's actually a win for the cloud provider as well. Because they are then able to put more customers on less hardware. Then they pay less money for electricity and pay less money on hardware investments. This is just a repeat of the virtualization / Dockerization of workloads, just on much higher scale.
I don't really believe the suggestions conflict with cloud provider motives. If all cloud customers reduce their workload requirements, it's actually a win for the cloud provider as well. Because they are then able to put more customers on less hardware. Then they pay less money for electricity and pay less money on hardware investments. This is just a repeat of the virtualization / Dockerization of workloads, just on much higher scale.
The author makes the same argument; or at least points out there is truth to that argument.
Google at least claiming to buy 100% of their usage from renewable sources, they are pretty open that it doesn't mean that is what feeds their electrons at all times of the day though. Better than most traditional computing / on premises sources though.
https://www.google.com/about/datacenters/renewable/
https://www.google.com/about/datacenters/renewable/
Something that I have also heard is that as well as just purchasing renewable energy, Google is also offsetting the carbon from their entire operations. So e.g. they calculate the average employee commute carbon emissions, or the carbon emitted in the supply chain for getting the free sushi to employee's mouths etc and offset that too.
I like that if it is true. I can't find anything specific on their report (1) - it is kinda vague: they say they're using renewable energy and are carbon-neutral in the same sentence, so not sure if that is just energy or the entire operation.
1 - https://sustainability.google/reports/environmental-report-2...
I like that if it is true. I can't find anything specific on their report (1) - it is kinda vague: they say they're using renewable energy and are carbon-neutral in the same sentence, so not sure if that is just energy or the entire operation.
1 - https://sustainability.google/reports/environmental-report-2...
Relevant :
https://www.sciencedaily.com/releases/2019/05/190523111408.h...
"100% renewables doesn't equal zero-carbon energy, and the difference is growing"
"100% renewables doesn't equal zero-carbon energy, and the difference is growing"
I buy 100% wind power at home, but its pretty much a scam I think. Of course when there isn't any wind I get coal/gas/nuclear generated power. The 99% of people who dont care dont care and its just a way to get some extra charge to some people who try to be green. I keep doing it to try to send some signal but might give up soon.
I've wonderd about buying "eco power", it's not like electrons have markings on them that say "I came from a generator attached to a wind turbine!". It's just that the wind turbines put some Megawatts into the pool and their buyers took the same amount of Megawatts.
Don't know where you are, but green energy providers, at least in the UK, either have a relationship with the producer where they buy directly from them, rather than from the grid, or buy specific types of energy from the grid. It IS possible to buy energy from a specific source and energy is traded constantly during the day on the energy exchange.
Bulb for example have more information about where their energy comes from: https://bulb.co.uk/energy/
Bulb for example have more information about where their energy comes from: https://bulb.co.uk/energy/
The whole point is to send that signal, increasing demand for wind power, and increasing the budget for the company so they can invest in new turbines.
Ok but how about all the carbon removed by cloud computing? Less commutes necessary thanks to video conferencing tools and work from home, less cars on streets due do deliveries (disputable, depends on how the city is laid out I suppose)...
Possibly less energy consumption for the same cloud service thanks to shared hosting, on demand instances etc
Well more and more computing power in one place won't be carbon neutral. I disagree with this article. It's like writing that 100 meter building can be more efficient than 100 buildings. Yes it takes less space horizontally but it's more complex to power it and support people then make 100 same buildings and provide each with wind / solar power.
That's what decentralisation is about and maybe somewhere in future we start building self sustaining villages with own small networks that barely need to communicate with outside world instead of building giant networks that are inefficient. P2P FTW
Low Tech Magazine's piece "Why We Need a Speed Limit for the Internet: The energy use of the internet can only stop growing when energy sources run out, unless we impose self-chosen limits." https://solar.lowtechmagazine.com/2015/10/can-the-internet-r... covers this situation from a systemic perspective, realizing problems in meeting every energy need without considering alternatives.
When you consider the impact of "cloud computing", you need to consider the offsets as well.
Enterprise datacenters have much higher PUEs than a hyperscale datacenter, and the hyperscale folks have an incentive to use energy as efficiently as possible. Hyper-scale datacenters often tend to be placed in locales where space and access to electricity is better.
There is always alot of navel gazing about anything environmental, and it's easy to throw rocks at what's new, because the facts are known and easy to find. Ultimately, I don't think that what the author is suggesting that we all do (ie. developing with an eye on efficiency) could meaningfully happen without cloud computing, because the "old way" of making big capital purchases for infrastructure is too disconnected from the operational costs.
Enterprise datacenters have much higher PUEs than a hyperscale datacenter, and the hyperscale folks have an incentive to use energy as efficiently as possible. Hyper-scale datacenters often tend to be placed in locales where space and access to electricity is better.
There is always alot of navel gazing about anything environmental, and it's easy to throw rocks at what's new, because the facts are known and easy to find. Ultimately, I don't think that what the author is suggesting that we all do (ie. developing with an eye on efficiency) could meaningfully happen without cloud computing, because the "old way" of making big capital purchases for infrastructure is too disconnected from the operational costs.
Sure but I think in a way they’re a necessary evil.
I don’t think we’re gonna get to a futuristic carbon neutral world without a hell of a lot of processing power
I don’t think we’re gonna get to a futuristic carbon neutral world without a hell of a lot of processing power
Since we don't know for sure what the capacity utilization of say AWS is, we do not know if they are running physical servers which are lightly loaded or not.
It might be that AWS runs a tight ship or it might be that they have entire aisles of servers spun up but which run for only 1 hour per day.
It might be that AWS runs a tight ship or it might be that they have entire aisles of servers spun up but which run for only 1 hour per day.
I would think they would turn off computers they don't need, since that is a huge saving on power and cooling.
They presumably can predict when they need to be turned on again, ready for use.
They presumably can predict when they need to be turned on again, ready for use.
Data centers are more efficient than people managing that same amount in their own homes or office.
Also, "1%....might come from burning coal....if it did....it would result in 1.2 billion tonnes of CO2". "Might"? "If" it did? How is that unreal theory useful to anyone?
Also, "1%....might come from burning coal....if it did....it would result in 1.2 billion tonnes of CO2". "Might"? "If" it did? How is that unreal theory useful to anyone?
Disclaimer: I'm the author of the article.
I'm not sure which part of it you believe is unreal, but I'll try to clarify; - the 1% electricity consumption of total world consumption is real - based on research done by IEA (https://www.iea.org/reports/tracking-buildings/data-centres-...) - The calculation on the effect of generating 200TWh of electricity from burning coal is 1.2 billion tonnes of CO2 emission is real - the link in the article points to the estimation - the fact that most of world's energy comes from fossile sources is real. While it's not 100% from coal, 67% is from fossile fuels; https://www.iea.org/reports/electricity-information-2019
And some countries are worse than others. https://edition.cnn.com/2019/09/10/asia/china-data-center-ca...
I'm not sure which part of it you believe is unreal, but I'll try to clarify; - the 1% electricity consumption of total world consumption is real - based on research done by IEA (https://www.iea.org/reports/tracking-buildings/data-centres-...) - The calculation on the effect of generating 200TWh of electricity from burning coal is 1.2 billion tonnes of CO2 emission is real - the link in the article points to the estimation - the fact that most of world's energy comes from fossile sources is real. While it's not 100% from coal, 67% is from fossile fuels; https://www.iea.org/reports/electricity-information-2019
And some countries are worse than others. https://edition.cnn.com/2019/09/10/asia/china-data-center-ca...
Thanks for clarifying
There is a lot of migration from smaller enterprise data centers to the cloud which may be net positive.
Reading the title, I was thinking this would be about cloud cover and not data centers to be honest…
proposed solutions:
> Increased energy efficiency on component level
> Increased energy efficiency on data center level
Why no “Increased efficiency on software level”? Is it simpler to throw hardware (and money, energy, carbon emissions) at the problem than find better solutions?
> Increased energy efficiency on component level
> Increased energy efficiency on data center level
Why no “Increased efficiency on software level”? Is it simpler to throw hardware (and money, energy, carbon emissions) at the problem than find better solutions?
Disclaimer: I'm the author of the piece.
If you read the very last paragraph, I actually propose increased efficiency on software level as a way to lower energy consumption. The two other (component and data center level) are things us as developers usually have little or no control over, they were merely used to illustrate that the industry as a whole is working on the problem, and that you as a developer actually can contribute.
If you read the very last paragraph, I actually propose increased efficiency on software level as a way to lower energy consumption. The two other (component and data center level) are things us as developers usually have little or no control over, they were merely used to illustrate that the industry as a whole is working on the problem, and that you as a developer actually can contribute.
Thanks for clarification and for the article! First time I only skimmed over headings and missed that part.
Here’s the number to keep in mind. A medium-sized cloud datacenter uses about as much power as a single passenger airliner, but emits no carbon. That’s your comparison.
Would be great if cloud providers added how much energy a service/server/etc is consuming and where that energy comes from in the dashboard.
This is dumb, the environmental impact is the same as any other electrical appliance but we don't ask "What is the environmental impact of life support?" because it doesn't matter.
The problem and solution have been clear for over a decade we all know it, we need clean renewable electricity.
edit: I guess if you are releasing 12 articles a day they can't all be insightful.
The problem and solution have been clear for over a decade we all know it, we need clean renewable electricity.
edit: I guess if you are releasing 12 articles a day they can't all be insightful.
Disclaimer: I'm the author of the article
I'm sorry to hear that you didn't like the article. I believe there is a fundamental difference between data centers and other electrical appliances because data centers consume energy 24/7/365, while electrical appliances do not.
It is a well known fact that electricity consumption from other sources, like heating or cooling buildings is much larger in comparison. Or even data networks and internet traffic in general; https://www.iea.org/reports/tracking-buildings/data-centres-... And there is a lot to do in those fields as well.
Unfortunately, renewables to overtake fossile are far away. Even with the development the last decade, wind and solar still count for only a fraction of world production; https://www.iea.org/reports/electricity-information-2019) Until then, we as developers can actually make a small contribution by offsetting getting the consumption from this particular source higher than it is today, even if number of data centers is exponentially increasing.
Hope you found some of the other articles insightful.
I'm sorry to hear that you didn't like the article. I believe there is a fundamental difference between data centers and other electrical appliances because data centers consume energy 24/7/365, while electrical appliances do not.
It is a well known fact that electricity consumption from other sources, like heating or cooling buildings is much larger in comparison. Or even data networks and internet traffic in general; https://www.iea.org/reports/tracking-buildings/data-centres-... And there is a lot to do in those fields as well.
Unfortunately, renewables to overtake fossile are far away. Even with the development the last decade, wind and solar still count for only a fraction of world production; https://www.iea.org/reports/electricity-information-2019) Until then, we as developers can actually make a small contribution by offsetting getting the consumption from this particular source higher than it is today, even if number of data centers is exponentially increasing.
Hope you found some of the other articles insightful.
> because data centers consume energy 24/7/365, while electrical appliances do not.
But life support isn't? It really doesn't matter that its on 24/7, the source of the energy is the problem not where its going.
In my country renewables are 50% of our electricity and growing.
The cost of installing a new windmill will be a lot cheaper than hiring a team of coders to optimise a single piece of code. You also have the benefit of solving the problem instead of the symptom. Again this is very dumb sorry.
But life support isn't? It really doesn't matter that its on 24/7, the source of the energy is the problem not where its going.
In my country renewables are 50% of our electricity and growing.
The cost of installing a new windmill will be a lot cheaper than hiring a team of coders to optimise a single piece of code. You also have the benefit of solving the problem instead of the symptom. Again this is very dumb sorry.
A lot of people seem to cling to this number, 3-5 year lifecycle for servers. But, much to the disappointment of software developers, machines in large clouds are often in service for 8-10 years. When you build your own computers there’s nobody to tell you that the warranty has ended, so you just leave them in the rack as long as they are TCO-positive.
I tried to share this link with someone on Facebook Messenger and it said it could not be shared due to violating their community standards. Only thing I can guess is it's something to do with the .christmas tld
I once read the suggestion to build a powerful cloud datacenter in orbital space, powering it with solar panels. That would solve any overheating problem too.
That would work for tasks that are not latency sensitive (i.e. long compute tasks, not interactive ones) and that don't require oodles of bandwidth (or aren't time sensitive so it doesn't matter that getting the data up there for processing takes a while).
IIRC typical packet round trip times for generic networking a single satellite is at least half a second, two or three times that being typical RTTs, and bandwidth is going to be orders of magnitude lower than can be achieved with land/sea wire based networks.
Also dissipating heat is more problematical in space than most people intuit do to vacuum conditions.
And getting the kit up there in the first place will have a large environmental impact, and upgrades will be somewhat more time consuming and otherwise resource intensive.
IIRC typical packet round trip times for generic networking a single satellite is at least half a second, two or three times that being typical RTTs, and bandwidth is going to be orders of magnitude lower than can be achieved with land/sea wire based networks.
Also dissipating heat is more problematical in space than most people intuit do to vacuum conditions.
And getting the kit up there in the first place will have a large environmental impact, and upgrades will be somewhat more time consuming and otherwise resource intensive.
Space might be nominally cold, but it's also a vacuum. It's rather difficult to dissipate heat up there.
Traubenfuchs(2)
> Although many cloud providers have pledged to decarbonize their data centers, none have ditched fossil fuels entirely
Most datacenters in France are located close to nuclear power plants and only run on fossil fuels in case of a prolonged power outage. To me this is as carbon neutral as it gets.