💾 Archived View for gemini.patatas.ca › posts › slow-fedi.gmi captured on 2024-05-10 at 10:24:22. Gemini links have been rewritten to link to archived content
-=-=-=-=-=-=-
2024-04-26
The fediverse has so much potential. It's possible we can ditch the silos of Big Tech, and reimagine how we connect online, this time on our own terms. We have the opportunity to expand the online space for people to exist as their full selves, and promote dignity, safety, and agency.
And yet, so much of the network is still shaped by the logic of those same Big Tech firms, even though they don't directly control the experience: aesthetic expectations that ensure dependence on their infrastructure, energy and resource requirements that promote centralization via economies of scale, a level of data density and intensity that limits participation. And at the center of all of this: a default level of redundancy that is utterly unsustainable, environmentally speaking. We need to address that, pronto.
Building this blog and running a fediverse server have made me more aware of good design practices. I'll admit, I'd never thought too much about this stuff beyond "cool layout" or "nice fonts", other than when I could clearly tell that a website or app was broken. But as I began to learn a little HTML and CSS, I started noticing people talking about things like accessibility and inclusion in web design.
They were talking about things like accounting for keyboard-only navigation. Providing both light and dark themes, for people with different visual issues. Using inclusive input fields on forms - e.g. not assuming a gender binary, or, not rejecting names that are 'too short' or 'too long'.
Again though, I probably wouldn't have learned any of these things if I hadn't noticed people talking about them. To make a larger impact, we really need to think about systemic approaches too, rather than relying on chance. For instance: before learning any coding, I built a website for my crafts business on Squarespace, and later did the same with WordPress - but at no point did either platform ever say "Hey, here's why you need to use 'alt' tags on your images." How great would it be if they did?
When I think about the fediverse, particularly when it comes to the design and behaviour of microblogging software like Mastodon, I see some discussion of accessibility, but far less discussion around issues of environmental sustainability - also something that's probably best approached at a systemic level.
The tech industry overall is an environmental catastrophe. So it's not terribly surprising that a network which, by and large, seeks to recreate "The Big Tech Experience, But Decentralized" will also have sustainability problems. For now, the fediverse isn't comparable to things like Bitcoin mining or AI model training - but as mentioned, this network currently defaults to an extraordinarily high degree of redundancy.
This redundancy requires a lot of hardware and energy. And aside from the environmental problems with that, it means that servers become expensive to run - which makes the entire network less resilient, and more vulnerable. For example: just listen to the statements from the Meta folks on the 2024-04-22 episode of the Dot Social Podcast:
You barely have to read between the lines: Meta understands that many admins struggle to keep their instances financially viable, and reckon they can take advantage of this unsustainability, use it as a reason to insert themselves into the fediverse at an infrastructural and protocol development level.
What's more, the financial cost of running a server creates a barrier to entry for people with low incomes, and/or who live in places with limited internet speeds and availability. The cost of managed hosting even for a single-user instance is way too expensive for most of the world. Here's Marco Rogers' experience:
$19/month is a *lot* of money if you live in a low-income country. Can we really claim to be an open, decentralized network if running a node is unrealistic for so many people?
Lastly: If we want any efforts toward environmental and economic sustainability to endure, we also have to reject the logics that have supported Big Tech's rise to dominance. So many folks expect a carbon copy of Twitter, Reddit, YouTube, or Instagram, without a proper examination of how and why those platforms got to where they are.
"When capitalism transforms something, it tends to add more machinery. The internet was no different. In the post-dot-com period, as firms began to find more promising paths to profitability, they also made the internet more complex. The simple static web page faded from view. In its place came the dynamic and interactive web application, designed to seize a user’s attention and stimulate their engagement, linked to elaborate subterranean systems of data collection and analysis.
"The online mall would be a computationally intensive affair. Just as the capitalist transformation of manufacturing meant replacing the workshop with the factory, the capitalist transformation of the internet would hatch factories of its own. These factories weren’t generally considered factories, though. That’s because they came to be known by a name that obscured their fundamentally industrial character: the cloud."
- Ben Tarnoff, Internet For The People
These 'Cloud Empires', as Wili Lehdonvirta calls them, are as massive and powerful as they are today in large part through the use of techniques eerily reminiscent of digital slot machines, as described here by Natasha Dow Schull, who coined the term 'ludic loop' to describe the hypnotic flow state their designers aimed to induce:
"They don’t talk about competition or excitement,” says Robert Hunter, the clinical director of the Problem Gambling Center in Las Vegas. “They talk about climbing into the screen and getting lost.” They are after “time on device,” to use the gambling industry’s term for a mode of machine gambling that is less about risk and excitement than about maintaining a hypnotic flow of action – a mode that is especially profitable for casinos."
- Natasha Dow Schull, "Slot Machines Are Designed To Addict", NYT Opinion, Oct 10, 2013 (footnote 1)
Now to be fair, someone scrolling Facebook isn't literally pumping quarters into a one-armed bandit. But as we all know, what they *are* providing is a steady stream of data to the platform.
Facebook tells advertisers that this data enables highly targeted ad campaigns, and the advertisers give them money - which in turn pays for more infrastructure, delivering more addictive & energy-intensive experiences, and increased data collection and analysis.
More broadly: The tech giants' increasingly centralized control over digital networks' means of production - the 'cloud factory' - couldn't have been achieved without inventing and inducing a corresponding means of consumption. We're forever being tempted into wanting higher resolutions, framerates, bitrates, and megapixel count, more immersive, 'lifelike' experiences - which, incidentally, often also means needing yet another new device, something that can handle that density of data. (footnote 2)
By bootstrapping demand for the services of their cloud factories, the tech giants widen the cost-of-entry moat around their dominant market positions; by using those dominant positions to mine our data, they fund further expansion of their computational, network, and storage infrastructure. If you're someone like Mark Zuckerberg or Sundar Pichai, this is a virtuous cycle. To the rest of us, it feels pretty vicious.
"[T]he key question: do we really need this much computation?"
- Paris Marx, Tech Won't Save Us, 2024/04/04
Not only has the tech industry concentrated an extraordinary amount of wealth and power in the hands of a tiny group of people, it's also become a major driver of the climate crisis. In 2019, well before the current AI hype, or the height of the Bitcoin mining boom, The Shift Project reported:
"Digital technologies now emit 4% of greenhouse gas emissions (GHG), that is to say more than civil aviation. This share could double from now to 2025 to reach 8% of all GHG emissions, i.e. the current share of car emissions. Reducing the threat of climate change requires drastically reducing global greenhouse gas emissions in the next few years; however, the energy consumption required for digital technologies is increasing by 9% a year."
These numbers are startling given how much attention gets (rightfully) paid to the emissions from airplanes and cars, versus how little we talk about the impact of digital technology. They're also surprising because computing has seen incredible efficiency gains over the last few decades. Even knowing that, you might still expect the energy numbers to rise a bit, as more of the world gains access to the Internet. But we're not seeing just a slow rise in energy usage, let alone a decline.
I guess we can chalk it up to the Jevons Paradox? (The short version: technological advances that improve the efficiency of a process tend to result in *greater* overall resource consumption, rather than less.)
Meaning that, if the cost of transmitting and storing data goes down by half, we'll end up transmitting and storing more than double what we used to.
Now, part of this is simply due to us shifting more of our methods over to digital, as it becomes possible to do so. We might send an email instead of a letter or a phone call, or opt for videoconferencing instead of meeting in person. And this can be a net benefit, environmentally speaking!
But I think the tech critics are right when they point out that a lot of our consumption comes down to a mindset - perpetuated by Big Tech, of course - that the digital has little to no impact on the ecological. So, something that could have been a conversation tomorrow becomes a dozen emails back and forth this evening; what could have been a conference call is now a videoconferencing session. Our phones fire off a copy of every picture we take to a server farm somewhere, and our archive of emails, documents, images, and videos spanning two decades sits indexed and searchable, ready for us at a moment's notice, on services so redundant that they boast about the number of 9's in their availability percentage.
It's tempting to look at the problem and assign the blame to our habits as individuals - scrolling social media, binge-watching TV shows, generally succumbing to the lure of the hypnotic ludic-loop rectangles and their dancing glow of light - and argue for a kind of digital ascetism. And not just for the sake of the planet - many people have been warning for years about the psychological harms of device addiction, and advocating for more mindful and minimalist ways to exist online.
And there's probably *some* truth there. I mean, I do think we need to cut our consumption, and recalibrate our expectations when it comes to our digital experiences. We definitely need to stop buying so many new devices. But is it fair to expect everybody to instantly overhaul their online lives so that they match one particular vision of what 'good' habits look like? We shouldn't forget: that 'mindless' scroll can be a genuinely therapeutic way to unwind for a little while, for someone who's chronically stressed and overworked. Maybe you work remotely - it's got to be nice to see your colleagues' faces every so often on video calls, even though in the aggregate, turning everyone's cameras off is way better for the environment.
What's more, as Ben Tarnoff & Moira Weigel pointed out back in 2018, tech giants such as Facebook, aware of the public's mistrust around their platforms' manipulative and addictive design, have co-opted the language of 'humane technology', and claim to be focused on providing more "meaningful" user experiences.
They're still manipulating users, of course, only this time nudging them toward the Facebook-constructed definition of 'meaningful' and 'humane' interactions - which (purely coincidentally!) also happen to be the sort that Facebook can data-mine more efficiently. They get the regulators and the public off their back, but stay firmly in control.
"We change as we change our tools, and our tools change us. But even though our continuous co-evolution with our machines is inevitable, the way it unfolds is not. Rather, it is determined by who owns and runs those machines. It is a question of power ..."
"Rather than trying to humanise technology, then, we should be trying to democratise it. We should be demanding that society as a whole gets to decide how we live with technology – rather than the small group of people who have captured society’s wealth."
- Ben Tarnoff & Moira Weigel, Why Silicon Valley Can't Fix Itself
One person might be a photographer whose main source of income is selling prints. Another may simply find joy in scrolling through #caturday posts - there's no accounting for taste! Someone else might have a low mobile data cap, an old or slow device, a low-powered server with limited storage, or prefers a mostly text-based experience. None of these are the "wrong" way to exist online.
And that's ok, because the promise, or so I was told, of a decentralized, federated, protocol-based network is that it puts those choices back in all of our hands. But so much of this network seems like it's built on the exact same logic as Big Tech's ludic-loop-enabling, high-density, high-availability, energy-intensive silos.
Like with accessible web design, it was mostly by chance that I started noticing things. One day, I got curious: how much data was being sent to and from my single-user Mastodon server? I was following around 650 accounts, and maybe a half-dozen hashtags on fedi-buzz relays. I figured a couple hundred megabytes was probably in the ballpark. And then I did some arithmetic. Transfer rates averaged around 50-60KB/sec - which works out to 150GB per month. That's five gigs every day - way out of proportion to what I felt like my usage was.
In case anyone's wondering if my server is an outlier, it matches the experience of another single-user instance, whose S3 storage held almost 180GB in media after a month:
https://www.micahwalter.com/how-much-ive-spent-so-far-running-my-own-mastodon-server-on-aws/
I get that 150 or 180 gigs might not sound like much, at least not to people where I live. But not too long ago it would have easily maxed out my ISP's data cap. Even now, for a lot of the world, running a server like this is an expensive proposition. And the most frustrating part about it?
Most of that data is media attachments that I may never even see.
Did I mention that most of this data is media attachments that I'll likely never even see?
Here's an example of how Mastodon (and most other fediverse microblogging software) works: I can attach up to four images up to 16MB each to a post (footnote 4), and when I hit send, the text of my post, along with some links and metadata, is sent to every server with at least one account that follows me. The links tell the remote servers where to find the four images, plus a reduced size version of each. The remote servers all automatically send requests to my server to download those 8 images. And my server (assuming it can handle all this!) happily obliges.
We're fully immersed in Big Tech's production-consumption logic. The remote servers don't know if the account that follows me is active, or if it's been dormant for months. They don't know if the person who owns the account is asleep or awake. Everything is pre-fetched for every follower so that, if at some point they happen to scroll their feeds all the way to my post, the images load a bit faster.
"How can we explore ways of reconnecting the digital to the physical, considering critical finite resources such as energy, minerals and water as well as the limited planetary sinks to absorb the accumulating waste. This shouldn’t be about shaming users or burdening them with the responsibility. The responsibility for reducing carbon and energy use in the digital world should start with those of us designing and building digital products and services. The internet is a vital tool and digital technology will be an important part of communicating, organising and reworking the future, so we need to make it fit for a sustainable world."
- Tom Jarrett
To reiterate: I'm not trying to yuck anyone's yums here. Plenty of folks want a snappy experience, with lots of high-res images, video clips, and so on, I get it. Degrowth should not imply austerity. But shouldn't there *also* be some easily-configured options for folks with slower devices and connections? For people that want to run the lightest, lowest-impact server they can - or who otherwise can't afford the expense of running one?
It's a downside of decentralization, especially when it's not accompanied by democratization: the impact of someone's design choices will be different depending on what your situation is. Many effects can go unseen by those making the choice, and sadly, it's often difficult to convince people they should care about things they can't see, or that don't directly affect them. Especially when they can say, hey, you're free to go use someone else's software, or, hey, go code it yourself.
"I think there's loads of good reasons to be optimistic about the path to designing more sustainable digital products and services. The good news is there's so much low-hanging fruit, so much wasteful redundancy built into internet infrastructure. And we can make a massive difference by simply *trying*. Because at the moment we're not even doing that, we're not even trying."
- Tom Jarrett, on the Green IO podcast
The default assumption is that admins will use S3 object storage to handle all these media files. I asked around - S3 isn't expensive compared to the cost of a VPS. But it does still cost money, creates reliance on larger entities, requires additional technical knowledge, and (usually) means that data is stored at a greater distance from users, which adds to the energy used for transmission. Plus there's the embodied emissions from all the drives that the data sits on, the energy used to cool the server racks, and the water consumption.
(Did I mention that I'll probably never even see most of the media that's sent to my server?)
If you don't have unlimited disk space, the alternative to hoarding everything in the cloud is to regularly purge older media from your server's cache. Unfortunately, the side effect of this is increasing overall data transmission, because avatars and headers will end up being fetched over and over again.
Maybe you're feeling like I'm nitpicking when it comes to avatars, but bear with me here. When I started getting curious about ways to lower my data footprint, I set uBlock Origin to block any media larger than 50KB on a website. If I want to load a blocked image anyway, I just click on the placeholder. It's been interesting to see which websites are putting in the work to compress images and minimize data transfer. But it was a bit of a shock to load up Mastodon's web view, and see uBlock blocking a *lot* of avatars ... that's when I realized that all these apparently tiny images get stored at original size! (footnote 5)
If that doesn't rustle your jimmies, consider this: remember how Mastodon sends 2 copies of each image attachment, one larger, one smaller? I was curious about that filesize ratio, so I picked an attachment at random. I wish I was joking: the ratio was 1:1. Two copies of the exact same image were being automatically sent to who knows how many servers, and being stored on who knows how many drives and S3 buckets. I'm assuming the image in question was below some size threshold for getting compressed or whatever, but like, are we serious about any of this or not?
There are existing efforts to solve some of these issues, for instance Jortage is basically pooled S3 storage that de-duplicates files to keep its overall size lower:
https://www.fastly.com/blog/community-spotlight-una-thompson-is-making-the-fediverse-more-manageable
The server maps at fediverse.observer highlight servers running on 'green hosting providers':
Akkoma's server software takes a different approach by not caching remote media at all, by default. Attachments are downloaded only when a user is scrolling their feed (1st link). However this is somewhat controversial for various reasons (2nd link).
https://docs.akkoma.dev/stable/configuration/storing_remote_media/
https://docs.gotosocial.org/en/latest/admin/media_caching/
Here are a few feature suggestions that I'd like to contribute to the overall discussion:
What I'd really love to see is movement from at least some fedi devs on this. My personal feeling is that GoToSocial is a great fit for this type of effort! And it would be great for groups like the SWICG to make environmental sustainability something that gets regularly put on the agenda.
As I said right at the top: the fediverse has so much potential. Making it the environmentally conscious network, as well as a truly globally accessible one, would go a long way toward showing the world why the social web is the future.
Hoping this gets more discussion rolling. Here's to a future a fediverse that's green, equitable, and independent!
1. If this kind of thing is your jam, podcast fans might want to give this fantastic episode a listen, in which Prof. Anna Kornbluh describes the 'synthetic immediacy' trap that pervades contemporary culture:
Novara Media Podcast 2024/03/21: Critical Theorists Hate This One Weird Trick
2. Likewise, our new devices can end up driving consumption. Would websites have so many enormous images, if Retina displays didn't exist?
Also related: I believe the AI push can be understood simply as the latest attempt to bootstrap demand for capital-intensive infrastructure. By throwing anything and everything against the wall to see what sticks, and cramming AI into everything they can, Big Tech clearly hopes to find something that people will start to rely on. If they somehow succeed, then the enormous infrastructural expense required to build and train a competing AI model means that the cost-of-entry moat protecting their market dominance has become even wider.
3. And if you take the lower number from earlier - 32g CO2e per gigabyte - and multiply that out over the course of a year, you get about 57kg CO2e, which is over 1% of the roughly 5 tonne per-capita emissions globally in 2022. To be fair, since 2021 there have continued to be efficiency gains in digital networks, and there are more renewables in the mix. On the other hand, as already noted, there are also other estimates of the CO2 equivalent of data transmission that peg it at around 400g CO2e/GB.
4. I got curious about this too, so I did a little test. Fired up the ol' image-editing software, created a 3840x2160 image, and saved it as a PNG file with no compression. The filesize is 23.8 MB. Saving the same image as a JPEG at 90% quality knocks over 50% off the file size. And that's not even mentioning the fact that the JPEG image is absolutely indistinguishable from the PNG to my eyes, or the reality that both have many times more pixels than any screen I own can even display. (update: exporting the image in WEBP format was nearly 75% smaller than PNG)
5. My rough estimate is that by not pushing header images to remote servers unless a user's profile is actually opened, and limiting avatar sizes to <10KB or so, around 5-10% of my server's overall data transfer could immediately be saved. Low hanging fruit!
6. See also: Tom Jarrett's "Lower Energy Instagram" concept - which I swear I hadn't actually seen when I first thought of the click-to-cache idea, despite the similarity! But credit where it's due: the idea to display alt-text in the image placeholder is all Tom's.
https://www.versobooks.com/en-ca/products/2674-internet-for-the-people
https://medium.com/interactive-designers-cookbook/the-lucid-truth-of-ludic-loops-caec3ad272da
"Slot Machines Are Designed To Addict", by Natasha Dow Schull (archive link)
https://en.wikipedia.org/wiki/Jevons_paradox
https://www.theguardian.com/news/2018/may/03/why-silicon-valley-cant-fix-itself-tech-humanism
https://danluu.com/slow-device/
https://docs.joinmastodon.org/user/posting/#media
Branch Magazine (ClimateAction.tech)
The Environmental Footprint of Social Media Hosting: Tinkering With Mastodon. EASST, 2022