the environmental impact of cloud computing
This commit is contained in:
parent
c84ac6405f
commit
0b67c46448
|
|
@ -0,0 +1,62 @@
|
|||
---
|
||||
title: "The Environmental Impact of Cloud Computing"
|
||||
date: 2024-02-25T20:23:00+01:00
|
||||
categories:
|
||||
- software
|
||||
tags:
|
||||
- AI
|
||||
- music
|
||||
---
|
||||
|
||||
Ana Rodrigues [too deleted her Spotify account](https://ohhelloana.blog/goodbye-spotify/), citing numerous valid reasons (see my [You Shouldn't Listen To Spotify](/post/2021/02/you-shouldnt-use-spotify) post), of which one in particular stood out for me: **Spotify is bad for the environment**. She cites Natalia Waniczek's [FFConf 2022 talk](https://ffconf.org/talks/2022_lil_natw_talk/) where Natalia presented painful facts such as:
|
||||
|
||||
1. If you were to listen to an album 27 times back to back, the printed CD would be more environmental friendly;
|
||||
2. Streaming Spotify for a working month emits as much CO2 per month as driving `32 km` by an average vehicle.
|
||||
|
||||
I wonder how we can achieve what Natalia's talk was all about---_working towards a greener world from behind the keyboard_, or, in other words, _frugal computing_. I mean, I know what the concept is about, and it's quite obvious what needs to be done (less!), but in a world where everything needs to be upscaled instead of downscaled, who's going to listen to the few voices swimming against the current?
|
||||
|
||||
At my current client, we're leveraging Kubernetes' isolated throwaway-pod system to auto-upscale and deploy clusters. Distributed---that is, _cloud-based_---enterprise solutions are without a single doubt less environmental friendly than their classic client-server counterparts: modern software is often upscaled too fast, and mordern software is needlessly complex/distributed across multiple energy consumers. Environmentalism apparently is never a key decisive part of the engineering puzzle.
|
||||
|
||||
The most conspicuous example is of course (yet again) generative AI. A few days ago, Nature published an article proclaiming that [Generative AI’s environmental costs are soaring---and mostly secret](https://www.nature.com/articles/d41586-024-00478-x):
|
||||
|
||||
> [...] Artificial intelligence (AI) industry is heading for an energy crisis. It’s an unusual admission. At the World Economic Forum’s annual meeting in Davos, Switzerland, Altman warned that the next wave of generative AI systems will consume vastly more power than expected, and that energy systems will struggle to cope. “There’s no way to get there without a breakthrough,” [said OpenAI chief executive Sam Altman].
|
||||
|
||||
Their solution? Certainly not downscaling. Instead, Altman invests heavily into nuclear fusion companies hoping for a breakthrough in the field of energy generation, not in the field of AI. That alone blew my mind. The article also contains a sobering message for people relying on these Large Language Models to find stuff instead of relying on good ol' internet searches:
|
||||
|
||||
> ChatGPT, the chatbot created by OpenAI in San Francisco, California, is already consuming the energy of 33,000 homes. It’s estimated that a search driven by generative AI uses four to five times the energy of a conventional web search. Within years, large AI systems are likely to need as much energy as entire nations.
|
||||
|
||||
Every time you ask ChatGPT something you could easily have found yourself using conventional methods, you're burning through five times as much energy. Considering my students nowadays use ChatGPT almost exclusively, that's insane. Of course OpenAI doesn't really want you to know this, and even if you did, what else would you do but shrug and use the system anyway?
|
||||
|
||||
This isn't just a political finger-pointing game; multiple peer-reviewed scientific publications have been warning us about the environmental impact of these systems for years; yet we chose to ignore them. I'm afraid one recent Nature article won't exactly trigger a change of heart either. And no, these "optimized" and "shared" systems are far from optimal: once your data reaches your device, it has passed through multiple data center(s), the core network, edge network, access network, ...
|
||||
|
||||
Just to name a few more:
|
||||
|
||||
- [Environmental Impacts of Shifting from Movie Disc Media to Movie Streaming: Case Study and Sensitivity Analysis](https://www.sciencedirect.com/science/article/pii/S2212827119301040)
|
||||
|
||||
> [...] The comparative LCA showed that, for the example of global warming potential, the environmental emissions of digital distribution are ultimately higher if the digital distribution leads to more than 4x the amount of movie viewing.
|
||||
|
||||
- [A Systematic Review of the Pros and Cons of Digital Pollution and its
|
||||
Impact on the Environment](https://www.researchgate.net/profile/Kyaw-Than-Oo/publication/369168632_A_Systematic_Review_of_the_Pros_and_Cons_of_Digital_Pollution_and_its_Impact_on_the_Environment/links/640c9b9366f8522c389966f2/A-Systematic-Review-of-the-Pros-and-Cons-of-Digital-Pollution-and-its-Impact-on-the-Environment.pdf)
|
||||
|
||||
The researchers summarized the total carbon footprint of the most popular applications by watching hours, and three different methods of estimating that footprint:
|
||||
|
||||
| Vid. Activity | Watch hours/day | avg. CO2/hour |
|
||||
|----------|-----------------|----------|
|
||||
| YouTube | `1 B` hours | `162.33 g` |
|
||||
| Netflix | `6 B `hours | `974.33 g` |
|
||||
| Facebook | `0.1 B` hours | `1623 g` |
|
||||
| TikTok | `2 B` hours | `243.83 g` |
|
||||
|
||||
|
||||
The problem is that as these giants constantly grow, this is historical or _stale_ data, thus accurate forecasting becomes more and more important.
|
||||
|
||||
- [Calculating the Carbon Footprint of Streaming Media:
|
||||
Beyond the Myth of Efficiency](https://computingwithinlimits.org/2022/papers/limits22-final-Makonin.pdf)
|
||||
|
||||
> Our model finds that streaming one hour of Netflix videos is roughly between 0.783–0.983 kWh of energy consumption depending on the number of simultaneous streams (10,000–50,000) being streamed from the data center. [...] Two interesting comparisons are that one hour of Netflix streaming is equivalent to about 1 kg of CO2 or the burning of 0.17–0.21 kg of coal.
|
||||
|
||||
The researchers had difficulties interpolating energy data and end with an open call:
|
||||
|
||||
> We call on industry (telecom, hosting, and streaming providers) to publish more factual numbers---not to remain secretive and silent. Allow researchers to verify and understand the impact of streaming on the environment.
|
||||
|
||||
Another hint at the concealment, number fiddling, and secrecy these large streaming providers tamper with, further clouding (ha!) the accurate calculation of the environmental impact of cloud computing. All these findings suggest it's high time not just for transparency, but also for responsible digital downscaling.
|
||||
|
|
@ -15,7 +15,7 @@ I have [sixteen analog notebooks](/post/2023/03/creating-journals-that-last/) di
|
|||
|
||||
Are you thinking what I'm thinking, Tony Stark?
|
||||
|
||||
|
||||
|
||||
|
||||
> Jarvis, which blog post ideas in the past month aren't written yet? Jarvis, what goes in my secret gingerbread spice mix again? Jarvis, what were my thoughts on project x again?
|
||||
|
||||
|
|
|
|||
Loading…
Reference in New Issue