Rhizome Climate Impact Portal

Measuring and understanding digital emissions is becoming a popular sub-topic of sustainability in the arts and culture sector. So much of the work we do in the digital art community and within the art community at large, relies on digital tools and applications that utilize cloud servers, online archives, web hosting, and more. As these tools become increasingly popular in recent years, there is also a growing interest in measuring the carbon emissions produced by these activities. But measuring digital emissions is not as straightforward as measuring emissions from commuting to work in a car, or from shipping artworks around the world. Digital emission tracking is still in its early stages of development and represents a space for growth for most organizations. 

Our Climate Impact Portal was created as a way to begin exploring this field and to ideally provide a series of tools, references, and examples to help grow this area of emissions tracking for the arts and culture sector. The Portal grew out of a nearly year-long research and coaching process supported by Teiger Foundation’s Climate Action Program Pilot, along with consulting from the Green Web Foundation. 

It’s important to note that calculating the climate impact of digital services is complex and imperfect. There are myriad elements that factor into the ultimate impact of a service. To take this website as an example, the code that produced this page is running on a proportioned part of a shared server that our service provider manages and that we have limited visibility into. We have to rely on formulas created by climate researchers to estimate based on our billing and usage data from the service provider instead of taking direct observations. After the server produced this page, the data had to travel over various internet nodes from where the server is geographically located in the eastern US to reach your browser. Finally, the type of device and network you’re using to view this page can have different impacts, for instance a high performance gaming pc vs a tablet or using a cell network can be more resource intensive than a wired connection. Due to these complexities, Rhizome decided to focus on our server infrastructure as a starting point to get the most accurate assessment of a significant area of our impact.

We also decided to limit our scope to CO₂ as the primary contributor of our impact, but there are also many other considerations that factor into a more complete assessment–such as water usage, electric grid demand and generation carbon intensity, along with other factors we’ll delve into as we expand on our calculation work below.

The information shared in this portal was informed by a survey that was conducted in the fall of 2023, which asked arts and culture professionals across North America what they already understood about their climate impact and what types of resources or tools would be helpful. Our goal is that the final result of this process and information that we have compiled here—both from our own experience tracking digital emissions, and from the external resources we have shared below—will support other arts and culture industry professionals in their own journey’s of understanding, measuring, and working to reduce their climate impact. 

Finally, this portal is an ongoing project and will be updated as we continue to refine our approach.

Rhizome’s Infrastructure

Rhizome champions born-digital art and culture through commissions, exhibitions, scholarship, and digital preservation. Founded by artist Mark Tribe as an email discussion list including some of the first artists to work online, Rhizome has played an integral role in the history of contemporary art engaged with digital technologies and the internet.

To support these digital preservation and artistic program efforts, Rhizome operates an expansive set of software and servers. This complex web of infrastructure is one of the motivators for initiating our climate impact work. There’s a balance between reliability and sustainability, both financially and with regards to resource usage, and engaging in this audit helped us better understand how to provision and optimize our infrastructure. Below is a quick overview of the services we steward and an interactive map of their requisite infrastructure.

First launched in 2016 as Webrecorder.io, and later renamed in 2020, Conifer is Rhizome’s open source web archiving service. A boundary pushing tool to capture, organize, and replay websites, Conifer requires an assortment of servers to store archived content, user data, and to support the complicated capturing and replay work of web archiving. 

Established in 1999, Rhizome’s ArtBase is an archive with over 2,200 artworks to date, primarily hosting works of net art and software, encompassing websites, games, browsers, and much more. Rhizome’s commitment to the preservation of works in the ArtBase has grown alongside the archive’s expansion in size, scope, and complexity over the years. Numerous works in Rhizome’s Artbase rely on Emulators to be accessible. These Emulators give access to legacy computer systems and are hosted on servers which require additional resources.

We also maintain video.rhizome.‎org, Rhizome’s self-hosted video platform driven by the fediverse service, PeerTube. PeerTube was created as an independent, peer-to-peer alternative to commercial video hosting platforms. Rhizome uses video.rhizome.‎org for the purpose of supporting access to artworks and as an institutional archive. Content includes migrated videos that had been previously published on other platforms that aren’t archive-friendly, including artist works and talks that were potentially at risk of erasure. Rhizome also uses this instance of PeerTube as a primary home for documenting ongoing programs and events.

Alongside these services is our primary site rhizome.org and various project sites and exhibitions such as Net Art Anthology or New Black Portraitures. These sites mostly run on a shared server or are hosted statically on a file server. All of our infrastructure is spread between Google Cloud and Amazon Web Services.

CO₂ Calculations

It bears repeating that climate impact assessment work is challenging. Fershad Irani, a consultant from the Green Web Foundation that Rhizome worked with, cued us into a helpful aphorism from George Box: “all models are wrong, but some are useful.” This nicely relates to what we were after–a working estimate of the impact of our services based on best approximations–but also raises an important caveat with this work. Due to the lack of direct access to the systems we use and their consumption data, we’re required to use averages, estimates, and best guesses to create our climate impact model. Utilizing these estimations gives us a relative baseline and provides feedback as we adjust and better optimize our infrastructure, allowing us to measure the reduction of our impact.

Service providers are slowly starting to integrate high level CO₂ estimates to their admin panels, but the data provided is not detailed enough to drill down and understand granularly what items are causing the biggest impact. It’s our hope that these estimates continue to mature and provide more accurate information.

To gather our baseline emissions we set up Cloud Carbon Footprint (CCF), a free open-source digital emissions measurement tool. CCF pulls billing and usage data from a service provider (currently supporting Amazon Web Services, Google Cloud, and Microsoft Azure) and applies an estimation methodology to produce a CO₂ emissions calculation.

The formula CCF uses to calculate emissions incorporates metrics like compute, storage, networking, and memory. 

Let’s look at the compute estimate as an example of how these formulas work:

Average Watts =
Min Watts + Avg vCPU Utilization * (Max Watts - Min Watts)

The average watts metric is the average power usage at a point in time. CCF uses constants for min and max watts based on public info for each service provider, and vCPU (virtual CPU) is either pulled from the service provider when available or a reasonable constant is used.

Compute Watt-Hours =
Average Watts * vCPU Hours

We can then apply the average energy usage (average watts) by the amount of time each of our services have been active (measured in hours) to calculate our total watt-hours.

Next we need to account for the efficiency of the data center our cloud provider runs. This value is called the Power Usage Effectiveness. The PUE is a number with a best possible value of 1, which would be a perfectly efficient system with no losses. CCF uses the values made publicly available by each service provider, for instance the PUE for Amazon Web Services is 1.135.

Finally, we’ll convert our usage estimate into kilowatt hours, apply the PUE value and the approximate amount of grid emissions for the region that the resources are located in to get our final estimate.

Server compute CO2 (grams) =
Compute Kilowatt-Hours * PUE * grid emissions factor (metric tons CO2 / kWh)

This is a quick high-level example of how these calculations are made. You can read in greater detail how each metric is being calculated in Cloud Carbon Footprint’s documentation.

To help put these numbers in perspective, let's look at a flight from New York City to Los Angeles. The distance traveled is approximately 2,500 miles which would emit around 600 lbs CO₂ per passenger. Or take a standard 22 MPG passenger car, which emits an estimated 27 lbs CO₂ per day on average.

Findings and Guidance

We’re incredibly grateful to have been granted the opportunity to spend time researching and learning about our climate impact and to step back and take a look at the entirety of our system. Below are some initial findings and the steps we are planning to implement to lower our climate impact.

Keep tabs on your infrastructure provisioning, size, and version. This is the most immediate and likely obvious step an organization can take, but something that can often be neglected with competing priorities of a small team. Cloud platforms have also made provisioning infrastructure so frictionless that it’s easy to set and forget something. Using tools like Cloud Carbon Footprint or Cloudcraft to track and map your infrastructure usage is incredibly helpful to ensure you’re only maintaining what you require. Finding a cadence that’s sustainable to review these systems, such as every quarter or new year, can also help. Additionally, cloud platforms often introduce more efficient and cost effective hardware. Keeping up to date can help lower costs (as these newer systems are often incentivized with lower cost of usage) and utilize your provider's most efficient hardware. It’s also important to keep an eye on the CPU or memory utilization of a machine and scale it to be appropriately sized. When Rhizome set up Cloud Carbon Footprint on our services we realized a database was running for an old project that was no longer online. Trimming anything unnecessary like this goes a long way.

Take into consideration where your servers are located. Most cloud providers offer various options where your server will be geographically located. This flexibility can be used in a couple ways to lower your impact. 

First, you can analyze the network traffic to your service. If it’s predominantly coming from a certain region, you can both save some resource usage and improve response times by locating the servers in that region. Similarly, if there is an equal amount of traffic between two disparate regions, you could consider provisioning two smaller servers to serve the respective regions instead of a single larger server that would require data to travel farther for some of the traffic. This calculus can become complex and is an area of active research.

Another consideration for where to geographically locate a server is the mix of power generation sources for that region’s power grid and the associated carbon emissions. There are organizations that collect and provide resources to determine the carbon intensity of the grid in different regions. Cloud Carbon Footprint includes a map, with a breakdown by service provider. There are some important caveats to this approach. In short, you should still strike a balance between where your traffic predominantly comes from; keep your provisioning relatively consistent and avoid chasing the lowest carbon intense region at any moment. These actions will allow your cloud provider to best utilize their server capacity and resources and forgo unintended effects on the grid operator. If there’s a region nearby your community with lower intensity, it can be worthwhile to locate there. 

Take steps to reduce the size of the server or compute resources required. As the graphs above illustrate, compute or CPU resources are the single biggest contributor to a server’s carbon emissions during operation. There are some techniques to reduce the amount of computing power your website or service will require. Caching is an effective strategy that stores aspects or whole pages and returns the cached value for a request instead of requiring the server to respond. This can allow the server to be much smaller while handling larger amounts of traffic. If a page changes frequently, using a cache that only lasts several seconds can still greatly reduce the server’s workload.

Other Modes: Rhizome’s Solar Protocol Server

Solar Protocol is a web platform hosted across a network of solar-powered servers set up in different locations around the world. A solar-powered server is a computer that is powered by a solar panel and a small battery. Each server can only offer intermittent connectivity that is dependent on available sunshine, the length of day and local weather conditions. When connected as a network, the servers coordinate to serve a website from whichever of them is enjoying the most sunshine at the time.

As part of our climate impact work, Rhizome is interested in exploring lower impact modes of connectivity. As our estimation work confirms, server computation is the most costly aspect of common hosting infrastructure. The Solar Protocol project combats this by creating a distributed, solar powered and low cost collection of servers to host a network. Rhizome’s node will be hosted at a community space in Los Angeles.

In conjunction with the launch of our new Solar Protocol node, Rhizome announced a microgrant call for artwork to be hosted on the Solar Protocol network. A launch event for the server and the commissioned works is forthcoming.

Acknowledgements

We would like to thank the following organizations and individuals for their support in the creation of the Climate Impact Portal:

Teiger Foundation supports contemporary visual arts with specific focus on the work of curators. As part of their support for contemporary visual arts, Teiger Foundation launched the Climate Action Pilot program in 2023 to support the implementation of climate action in the arts and culture sector. By pairing grantees with a climate coach, grantees receive tailored support and financial resources to realize their climate action goals. 

Rute Collaborative acts as the climate coach for the Teiger Foundation’s Climate Action Program. Following a holistic approach to sustainability, Rute Collaborative works with organizations and individuals from the arts and culture sector to create actionable and meaningful climate action plans, and to inspire organizational change.

Green Web Foundation is a nonprofit organization supporting and working toward the goal of a “fossil-free internet by 2030.” Green Web Foundation offers free tools and resources that help manage and measure the carbon emissions from digital applications and technologies, and supports leaders and policymakers in their work to advocate for a sustainable and just internet. Fershad Irani, a developer and consultant with Green Web Foundation, consulted on the development and build of the Climate Impact Portal. His support and expertise played a key role in making this project possible.

Laura Coombs is a designer in New York, working on visual identities and editorial projects with museums, artists, and architects. She teaches graphic design at Princeton University. Laura provided design guidance for the climate impact portal.

Sources

The following is a collection of helpful resources that we gathered and used during the creation of the Climate Impact Portal. This collection is not exhaustive, but offers some ways to get started learning about digital emissions and calculating your own digital emissions; for developers, how to build sustainable websites and other digital applications.