The environmental footprint of social media hosting: Tinkering with Mastodon

by Stefan Laser, Anne Pasek, Estrid Sørensen, Mél Hogan, Mace Ojala, Jens Fehrenbacher, Maximilian Gregor Hepach, Leman Çelik, Koushik Ravi Kumar

A growing body of literature on waste and discard studies has crafted a powerful critique of waste management and politics (Callén and Sánchez Criado 2015; Liboiron and Lepawsky 2022; Gille and Lepawsky 2022; Ek and Johansson 2020). In today’s dominant waste regime, waste is naturalized as a burden of individual consumers while focusing on the end-of-pipe of consumer waste. The largest share of most waste materials, however, is generated during the production process and due to the reliance on fossil-fuelled power grids (Stoekl 2007). For digital technologies, with their dependence on highly-processed mineral and metallic components, mining wastes can still be several orders of magnitude larger (Lepawsky 2018). Therefore, it’s increasingly clear to our eyes that economic structures that stabilize excessive consumption, extraction, and energy-intensive practices must be challenged and abolished. 

The example of electronic waste is a compelling one. Citizens of the world witness an ever-growing pile of hazardous materials and a complex stream of discarded electronics that are insufficiently addressed by sophisticated, yet shallow e-waste management policies (Lepawsky 2018). Faith in redemptive approaches or technology that purports to eradicate waste and adhere to an ideal of pure nature is also problematic. “Zero waste” and “net zero” are nicely packaged advertising (MacBride 2011), “greenwashing.” Not only will they not work, but in the quest for purity, impossible hopes are expressed, and harmful ideologies creep into material politics (Shotwell 2016; Balayannis and Garnett 2020). Yet it makes sense to think beyond the material streams of matter. Waste, here, can also mean emotional waste, time lost on apps, and the loss of commitment to communities that crumble when nourishing infrastructure is disrupted.

The waste generated by the digital media we consume is rendered invisible through the cloud metaphor. As J. R. Carpenter (2016) has argued, amongst many others, the “language of The Cloud is a barrier” to grappling with the actual material costs and impacts of internet infrastructures. While in the past, the so-called Twitter ‘fail whale’ (an icon that appeared during service outages) used to at least be a proxy for Twitter’s buckling infrastructure, present-day usage is generally without interruption. Or, rather, interruptions are harder to notice. We argue that the seamlessness of the user experience is illusionary, as it covers violent infrastructural practices of material extraction at the very foundation of technological infrastructure and its sustained maintenance.

Against this backdrop, we approach the problem of the environmental footprint of social media with some caution. We want to discuss ways to waste differently: by reflecting material investments on the infrastructural level, and looking for ways to make material entanglements transparent and ethically addressable.

The sudden influx of new Mastodon users in November 2022 has challenged the limits of Mastodon’s federated infrastructure. Mastodon now has roughly 8M users logging in through more than 10k serves. A month ago, it was a fraction of this. Following this increase in users, these have to be patient, as text and other media flow more slowly through the network. However, where Twitter users were only ever able to wait out outages, Mastodon users can balance the load on the infrastructure by creating new servers/instances for themselves, reminiscent of principles of peer-to-peer networking. In this process, novices to hosting web services have to face the material and elemental aspects of social media: as user numbers increase, so do the storage requirements and operating costs. With an economic model that is independent of online advertisement, those costs fall on system administrators and become visible when they request donations from users.

The occasion presents an opportunity to reflect on the waste we generate online and to imagine social media otherwise. It shows that there are costs to running services online, and that without advertisement-reliant economic models, those costs fall on system administrators and their supporters.

We are a group that has started to meet through virtual means and builds on local ties in Europe and Canada. Our expertise is spread across media studies and science and technology studies. While many of us were active users of “the bird site,” we all started to tinker with Mastodon after the controversial acquisition and dramatic reconfiguration of Twitter by billionaire businessman Elon Musk in late 2022.

The discussions on social media (fig. 1) were followed up on Zoom and are moderated by Stefan. We document our discussions with the help of Etherpad.

Our conversations have proceeded, in part, out of a shared sense of urgency (and we recognize that repeated manufacture of a shared sense of urgency is a key operation logic of social media, and a source of its toxicity). Instead of debating the digital public sphere or digital climate emissions as an abstract or remote research question, we are moved by an ethical and practical need to both secure a digital forum for intellectual exchange and social connection and respond to the ongoing climate crisis. Twitter’s apparent meltdown, in other words, is an environmental and interpersonal matter. What seemed solid is now a ruin.

Figure 1: Discussing environmental footprint on social media: Mastodon (left) and Twitter (right). Screenshots by Stefan Laser
Building shared fora on ruins

With 436 million active users, Twitter is one of the biggest social networks, yet small compared to YouTube (2.5 billion) or Facebook (2.9 billion, Statista). Calling it a ‘public sphere’ is thus to overstate many possible meanings of the word ‘public.’ Still, it is a politically relevant forum in which opinions are formed and developed, and it has become a good way for academics to network internationally. At their best, Twitter conversations and contacts have proven to be excellent sources for quickly spreading research results and insights as well as for intellectual and career networking, news about academia, recommendations and references for research projects and teaching, and calling out misconduct, inviting to conferences and other events, and collegial goodwill and camaraderie. For those of us who teach in smaller universities, and in smaller research fields for whom the closest collaborators and peers are often far away, and especially amid the interruptions to conference routines during the COVID-19 pandemic, the platform has become a meaningful way to maintain connections with distant and more diverse voices. This seems to be especially true for our colleagues of color (Okoye VO 2021). Additionally, as researchers look for lower-carbon alternatives to academic air travel, the platform has played a role in alternative forums for research exchange (DIY Methods 2022 Proceedings 2022).

It is anything but obvious that Mastodon can, will, or ought to provide a replacement for all these uses. For a start, let us take one step back. What kind of social media are we talking about here?

Mastodon is named after a long-extinct mammal species; the leading developer and creator Eugen Rochko admits that naming is not his strong point (see also Wray 2017 about mythologization, valorization and de-extinction of necrofauna). Mastodon has been around since 2016 and was created as an antagonistic alternative to Twitter during earlier platform anxiety (Zulli et al. 2020). 

Mastodon differs from commercial social media in four ways. First, Mastodon is not centrally organized. Instead, it takes the form of decentralized, free, open-source software (licensed through the GNU Affero General Public License 3.0) that anyone can host on a server. This creates a local microblogging ‘instance’, which may, in turn, federate with other instances. It is part of a larger universe or “fediverse” of other media (e.g., alternatives to YouTube or Instagram like PeerTube and Pixelfed) that communicate via the W3C open protocol ActivityPub (2017). Second, there is no algorithm to push engagement but only feeds and lists organized by time, like on early Twitter. The algorithms at hand are only used for certain services, like transitioning between servers without losing followers. Third, instead of relying on algorithmic power and machine learning magic, Mastodon developers have invested energy in crafting content moderation tools, various options to block harmful users and entire instances, and implemented gadgets such as content warning options that improve the user experience. Both, the tools and their genealogy matter. Scrolling through GitHub history and considering critical discussions on Mastodon, we learned that these tools grew out of community protests pushed by queer folks, protests partly against the main developer who does not always acknowledge critical input (see Jon Pincus (2022) partial history of Mastodon). Lastly, Mastodon is ad-free. The software development and individual instances are supported by funding and volunteer labor. Notwithstanding these four differences, for end-users, Mastodon is functionally and practically a drop-in replacement of familiar and hegemonic social media services; the exciting difference is precisely in the ownership and governance of the infrastructure.

As a result of the decentralized approach, there is a slightly higher barrier to join the network. There are new and diverse social norms and patchier outcomes for up- and response times. Like Twitter, users curate a network of friends and microblogged content within instances and broader “federated” servers (Zulli et al. 2020). What is absent on Mastodon is the ability to quote Tweet content and algorithmically organized news feeds, and so far, there is limited virality to it. 

Technical properties have social consequences. In this sense, the development of Mastodon is an exciting live experiment. Per Winner (1980; for a classic critique, see Woolgar 1991), the platform is more compatible with certain kinds of social relations and not others – less clout chasing, a greater culture of image descriptions and consent, etc. Mastodon is classified as an “alternative social media” (Zulli et al. 2020) because it takes core features of platforms such as Twitter and Reddit and applies them in a non-profit-oriented way, with the result that niche communities are nurtured, content moderation can be distributed and adapted depending on the community, and, via independent and open code development, server operators are given additional design freedom to go their own way beyond the main code (for example, via their own “forkes” in which longer posts are allowed or certain features are added).

The key point for our experiment is that the decentralized approach relies on users choosing a server to log in to; there is more than one place to go. At the level of user experience, this is intentional friction, a “seamful design” (Weiser 1994; Chalmers et al 2003). Anyone with the competencies, resources and infrastructure can run a server: companies and nonprofits, individuals and communities. Using existing hosting packages and containers, both peace lovers and war makers can set up a Mastodon instance based on online instructions. Yet how it will unfold from there on will vary. This is precisely what we as STS and (e-)waste scholars are interested in. 

A challenge for social media infrastructure is handling the quick and unpredictable changes in traffic. Scalability is the “value proposal” of cloud computing (Fehling et al. 2014). For example, we can predict bursts during sports events and elections. Yet, bursts also happen during environmental catastrophes, political protests and upheaval, content or influenza going viral, military attacks, etc. This cannot be foreseen. But platforms must calculate with this, and they regularly fail. For instance, although Twitter has based its service on combinations of cloud and data center solutions, and on controlling backbone and edge traffic (Hashemi 2017). It has a hidden history of outages during traffic bursts, in 2022 alone going down in January, March, July, and September (https://www.datacenterdynamics.com/en/news/twitter-musk-layoffs-outages-server-overheat/). 

On the one hand, it is questionable if a decentralized network such as Mastodon – both in terms of technology and staff infrastructure – can ensure continuous service. On the other hand, outages that make it impossible to traffic masses of panic posts may also be an advantage to healthier public debate and quality time online. We may revise our social media-shaped desires to be online and continuously produce and consume information. These networks are shaped and challenged by use. In return, the user experience is shaped and challenged by the viability, sustainability, and resilience of its infrastructure (and what that infrastructure relies on – like electricity and water.)

We started our online exchange with the question of whether Twitter consumes more energy and has a greater hunger for powerful and up-to-date hardware and to what extent the centralized servers and data centers partially offset this through energy efficiencies. One interlocutor used the Mastodon thread to hint at a comparison of Zoom/Microsoft Teams vs. Big Blue Button, a comparison of a commercial vs. an open system, and emphasizes the benefits of the open system. Unfortunately, there are no figures available on the energy consumption of Twitter. Absent such studies, we can only assume that the dependence on advertising and machine learning demands significantly outweighs the economics of scale on the commercial platform. So there has to be a surplus. Crucially, we do not trust claims of net zero emissions, especially considering the considerable savings Elon Musk wants to make on data centers.

Critical data center studies (Hogan et al. 2022) show that energy consumption is closely related to waste practices. The drive for ever-higher efficiency in data center management leads to the rapid replacement of hardware. In contrast, embodied carbon emissions and e-waste are largely ignored, i.e., not included in the calculation of standardized ratios of the industry. In addition, there is a strong separation of hardware and software so that operators do not ask how the need for targeted and nudge-oriented advertising (often powered by machine learning) drives up energy and computing requirements. Beyond this central focus on industrial practices, wasteful routines of users also emerge, such as dialing into mobile internet networks or producing high-resolution screens. It is important to emphasize that the consequences of such waste practices are not equally distributed but are borne in particular in the global South and by minorities (Laser/Schlitz 2019). By engaging with Mastodon, we argue that infrastructures can be re-arranged and managed differently.

What kinds of worlds are not probable but possible from the ruins of Twitter? Mastodon might not be the next big thing. Yet it is an exciting network that many people are experimenting with and, for STS scholars, offers entry points to learn through practical engagement. Perhaps more important than Mastodon per se is the idea of othernets (Dourish 2017 chapter 7); the internet we have is not a necessity, and might take a very different shape and feel different based on new collectivities. If Mastodon has a less devastating impact on the environment, what else about our internet can we change, or make a case for changing? 

Towards a sunnier social media

As open-source software, Mastodon lends itself to more-than-software-based experimentation. Our collective is interested in grappling with the practicalities and wasting practices of hosting Mastodon instances through solar energy; we desire a solarpunk Internet! This way, we follow an elemental approach orbiting around the sun. It is, in the words of Brain, Nathanson, and Piantella (2022), a form of energy-centred design that follows protocols and forms of ‘natural intelligence’ instead of ad-tech algorithms.

In practice, this would mean building, configuring, and maintaining servers on energy-efficient computers (such as Raspberry Pis) and an off-grid solar photovoltaic energy system. We plan to pilot this idea in Canada and Germany, creating a network that stores and serves data through the Mastodon software differently based on where the sun is shining (and, perhaps, powers down entirely in periods of darkness and inclement weather). This low-carbon method (Pasek and Piantella; Pasek 2020; see also Landa 2021; Landa and Riggelsen Gjørding 2021) would allow us to measure data flows, energy consumption and production, and relate them to user interactions and weather conditions. In this way, we can address our lack of knowledge about the energy consumption of social media by generating statistics ourselves, which can become the basis for discussions. Furthermore, this setup allows experimenting with ways of disclosure: Besides the documentation in texts like this one, a Mastodon bot is planned that feeds the data that the Raspberry Pi server digs up back into the discussion on Mastodon.

What is exciting and almost poetic about solar energy is that the energetic constraints affect the engagement of users in very different ways than the advertising-induced triggers of the commercial alternative: when the sun goes down, server’s power decreases and—depending on battery strength and activity—may go offline, resulting in environmentally-determined periods of rest. Situating matters. But it is strange. We noted that since the early days of homepages, many internet consumers could not or did not have to imagine themselves as server hosts until now, and indeed were alienated from this idea. Here, too, there is a rich ethical resource for reflection and alternative world-making. Last but not least, note that our group consists of scholars from both Europe and Canada. Due to the time-space difference of almost half a day, it took negotiation to find a slot for a teleconference. But for hosting social media through solar renewable energy, it is all the more exciting that we are in entirely different places rotating around the sun and receiving its radiation.

This is actual work in progress. The question of costs and how we relate our system to administrators will accompany us, as will the question of where and in what form we want to tap and store energy – if we wish to store it at all. For example, in Germany, this has to do with regulatory and aesthetic matters: at the campus in the Ruhr area, we have to maneuver the heritage protection of the historical campus, which is not immediately compatible with setting up solar panels outside our office windows.

Our research interests, desires, and practice, therefore, bring us to a very different set of scales, challenges, and responsive behaviors than those germane to Twitter. As a form of critical making (Ratto 2011), however, the project prototypes and prefigures many interesting dynamics with great potential to yield insights into the seemingly intractable sociotechnical problems of contemporary social media and its participation in professional academic life. As a form of critical making, it also involves our practical and affective investment in acts of exploratory discovery and maintenance. It seems to transform the problem of the scholarly commons into matters of both environmental and relational care (Puig de la Bellacasa 2017).

The notion of care invites us to think about what good forms of wasting may be. We assume that solar energy is an acceptable form of energy; we want to use equipment that we already possess (although gadgets had to be ordered, too); and we refrain from advertising to save computing capacity. But it immediately gets complicated. Setting up infrastructures is a care-full task involving questions of technology selection and division of labor, which at Mastodon extends up to the moderation team.

So what will be set up? We rely on an experimental and iterative approach. In other words, we are not starting with a large network for thousands of users, but first want to check whether and how we could get something to work in a small setup. Growing follows birthing. We use this opportunity to document and reflect on what kind of materials we are wasting and the worlds we establish and cherish. This qualitative research approach links methods from the social sciences and humanities with critical computing. We invite others to do so, too, working with local resources, needs, and visions. And consider contributing to the discussion on Mastodon!

Bibliography

Brain T, Nathanson A and Piantella B (2022) Solar Protocol: Exploring Energy-Centered Design. In: Eighth Workshop on Computing within Limits 2022, 21 June 2022, p. 7. Available at: https://computingwithinlimits.org/2022/papers/limits22-final-Brain.pdf.

Callén B and Sánchez Criado T. (2015). Vulnerability Tests. Matters of “Care for Matter” in E-Waste Practices. Tecnoscienza. Italian Journal of Science & Technology Studies 6(2). Available at: http://www.tecnoscienza.net/index.php/tsj/article/view/234

Carpenter, JR (2016) The Gathering Cloud. Available at: http://luckysoap.com/thegatheringcloud/frontispiece.html (accessed 28 November 2022).

M. Chalmers, I. MacColl and M. Bell, “Seamful design: showing the seams in wearable computing,” 2003 IEE Eurowearable, 2003, pp. 11-16, doi: 10.1049/ic:20030140.

Critical Studies of the Cloud / Data Center Studies (ongoing open bibliography): https://docs.google.com/document/d/1bnbDNTXlpddbfuShMgRKWvWLFIKUKBchUkluMMOVRkw/edit#heading=h.2bmuylggugur and https://www.criticalstudiesofthe.cloud/

DIY Methods 2022 Proceedings (2022) DIY Methods 2022 Conference Proceedings. In: 2022. Low-Carbon Research Methods Group. Available at: https://hcommons.org/deposits/item/hc:48563/ (accessed 25 November 2022).

Dourish P. (2017). The Stuff of Bits: An Essay on the Materialities of Information. MIT Press

Fehling C, Leymann F, Retter F, Schupeck W, Arbitter P (2014).  Cloud Computing Patterns. Fundamentals to Design, Build, and Manage Cloud Applications. Springer Vienna.

Hashemi, Mazdak (2017) The Infrastructure Behind Twitter: Scale. https://blog.twitter.com/engineering/en_us/topics/infrastructure/2017/the-infrastructure-behind-twitter-scale

Hogan M., Edwards D., & Cooper Z. G. T. (2022) 5 Things about Critical Data Center Studies. Commonplace. https://doi.org/10.21428/6ffd8432.af5934aa

Landa L (2021) Ecological Thought and Digital Technology What can be learned by following a “hands-on” approach to sustainable IT? ETHOS Lab blog Metadata. https://ethos.itu.dk/ecological-thought-and-digital-technology-what-can-be-learned-by-following-a-hands-on-approach-to-sustainable-it/

Landa L and Riggelsen Gjørding M (2021). Why Our Website Will Not Always be Accessible… on ETHOS Lab blog Metadata. Accessed on 29 November 2022). https://ethos.itu.dk/why-our-website-will-not-always-be-accessible/

Laser S., & Schlitz N. (2019). Facing Frictions: Waste and Globalised Inequalities. Journal for Development Studies, 35(2), 5–32. https://doi.org/10.20446/JEP-2414-3197-35-2-5

Lepawsky J (2018) Reassembling Rubbish: Worlding Electronic Waste. Cambridge, MIT Press. DOI: 10.7551/mitpress/11111.001.0001.

Gille Z, & Lepawsky J (Eds.) (2022) The Routledge handbook of waste studies. Routledge, Taylor & Francis Group.

Liboiron M, & Lepawsky J (2022). Discard studies: Wasting, systems, and power. http://mitpress.mit.edu/9780262543651

Okoye VO (2021) Black digital outer spaces: Constellations of relation and care on Twitter. Transactions of the Institute of British Geographers 46(4): 806–809. DOI: 10.1111/tran.12495.

Pasek A. (2020) Low-Carbon Research: Building a Greener and More Inclusive Academy  Engaging Science, Technology, and Society. https://doi.org/10.17351/ests2020.363

Pasek and Piantella. Low carbon methods. https://lowcarbonmethods.com/local/zine.html (failed to access on 29 November 2022)

Puig de la Bellacasa M (2017) Matters of Care: Speculative Ethics in More than Human Worlds. Minneapolis; London: University of Minnesota Press.

Pincus J (2022) Mastodon: a partial history (DRAFT). https://privacy.thenexus.today/mastodon-a-partial-history/ 

Weiser M (1994) Creating the invisible interface: (invited talk). ACM Conf on User Interface Software and Technology (UIST94).

W3C Social Web Working Group. ActivityHub (2017). https://www.w3.org/TR/activitypub/

Winner L (1980) Do Artifacts have Politics? Daedalus 109: 121–136.

Woolgar S (1991). “The Turn to Technology in Social Studies of Science.” Science, Technology, & Human Values 16, no. 1. 20–50. http://www.jstor.org/stable/690038.

Wray B. (2017). Rise of the Necrofauna: The Science, Ethics, and Risks of De-Extinction. Greystone books.

Zulli D, Liu M., & Gehl R. (2020) Rethinking the “social” in “social media”: Insights into topology, abstraction, and scale on the Mastodon social network. New Media & Society, 22(7), 1188–1205. https://doi.org/10.1177/1461444820912533