Over the course of 15 years, from 2002 to 2017, Internet traffic increased approximately 500-fold. The associatedCO2 emissions were estimated at 762 million metric tons in 2018.
So let’s imagine that, much like the Nutri-Score, we had a tool—to measure our environmental footprint when we click on a web page—that assigns a rating between A and G to the HTTP request. Let’s also suppose that these requests are archived, year after year, in a public database such as HTTP Archive. By regularly exploring this database, we could track changes in the environmental footprint of HTTP requests.
This is the goal of EcoIndex, founded in 2014, which refers both to a set of best practices for building a website and to a software tool that evaluates several factors for a given URL: its absolute environmental efficiency using a scoring system on a scale from 0 to 100 (the higher the score, the better); its relative ecological performance using a rating ranging from A to G, similar to what is used for household appliances or food products (Nutri-Score); the page’s technical footprint (size, complexity, etc.); and the associated environmental footprint (greenhouse gases emitted, water resources consumed).
Its goal is to help as many people as possible become aware of the environmental impact of HTTP requests and to offer concrete solutions to reduce it. While this model—based on the technical page metrics mentioned earlier—is fairly easy to understand, it also has its limitations.
Let’s first try to understand what this model entails. It’s important to note that estimating the carbon footprint of human activities cannot be done directly: the method used generally relies on a targeted activity model specific to the field being studied. This is the case with EcoIndex, which covers only HTTP requests and not all web activities. This metric is based on the “3-tier” concept, which considers three parameters: client, server, and network.
The “historical” version of EcoIndex consists of a browser plug-in that works as follows: the user enters a URL into EcoIndex, which then sends it to the server. The server returns an HTML page to the browser containing the responses to the request. The plug-in measures the application’s footprint in terms of the number of elements on the web page (the number of HTML tags, denoted as `dom`), the number of requests in the returned page (`requests`), and finally calculates the number of bytes in the returned HTML page (`size`) that traveled through the network.
These values are entered into the EcoIndex algorithm to measure performance and environmental impact.
An additional analysis by an expert is essential for a comprehensive and reliable operational assessment of environmental performance. This is because EcoIndex does not take into account the impact of the computer making the query or the user’s journey. Only a single query, isolated from any usage context, is analyzed—similar to the Nutri-Score or washing machines.
Similarly, when the request is processed on the server side in a data center (for example, at Google when the URL is http://www.google.com), EcoIndex does not take into account the environmental impact of that server in the traditional sense of life cycle assessments (LCA), nor that of the various network devices through which the request passes between the user’s device and the data center.
However, it allows us to discuss the models and their attributes that would significantly characterize the web’s environmental impact, as reduced to the scope of HTTP requests. Other advantages of EcoIndex are that the loading, creation, and display of the page in the browser are not simulated, and that the three parameters— dom, requests, and size —reflect an architecture that governs the macroscopic behavior of a web request; therefore, EcoIndex makes sense.
Under this system, environmental performance is calculated based on standardized, fixed values that are set once and for all and embedded in the model, without taking into account variations over time—for example, from one period to the next, such as lockdowns, holidays, etc.—or the user’s geographic location.
Furthermore, it is not the ` dom`, `requests`, and `size` parameters themselves that are considered, but rather values corresponding to quantiles—that is, a small number of values determined by retrieving these three parameters from the URLs in a reference URL database, HTTP Archive.
One might wonder about the long-term stability of these quantiles: Are they the same in 2023 as they were in 2020, when they were determined for the historical EcoIndex? In principle, websites are regularly updated to adopt best practices in eco-design over time—there is no reason why the quantiles should be set once and for all.
Another minor point: for certain websites—such as those of major media outlets—which are dynamic, the Ecoindex score is likely to change from day to day, though probably not too drastically—for example, by dropping from an A to a G.
In fact, even a dynamic website generally follows more or less the same template, consisting of editable elements (text, images, background, colors). You replace one piece of text with another, one image with another, without fundamentally changing anything… Given this phenomenon, EcoIndex seems robust to us because this “template” does not change.
However, the grades A–G correspond to EcoIndex ranges of 100–81 for A and 10–0 for G, without any clear explanation of what these ranges actually represent: How were these different thresholds determined? Are they equivalent to the quantiles for the HTTP Archive’s EcoIndex measurements? They are close but do not coincide exactly.
Finally, the historical model does not, at first glance, lend itself to the introduction of new attributes other than the three-tier structure into the model.
However, we could consider incorporating concepts related to the energy mix and proposing a new EcoIndex+ metric that assigns ratings approaching “A” for carbon-free energy used on both the client and server sides, and ratings around “G” if the energy sources involved are carbon-intensive. If the HTTP request goes through a 4G/5G mobile network, we could also factor in the mobile carrier’scarbon footprint, which would provide a more comprehensive view of the EcoIndex+.
To ensure a more comprehensive set of attributes to be incorporated into EcoIndex+, the community must first agree on these new criteria and then establish calculation methods capable of processing a large number of attributes using machine learning.
From the perspective of environmental impact metrics and best practices in eco-design for websites, EcoIndex is a simple approach that helps shed light on issues related to the role of digital technology in global warming. The indicator is particularly useful for improving subsequent versions of websites.
However, there is still work to be done to, on the one hand, deepen our understanding and better grasp the relationships between various high-level models—such as the three-tier architecture—and field analyses—such as the life cycle of a digital product or piece of equipment.
Furthermore, the initial model should be examined using data science approaches—that is, by exploring and analyzing the data to derive a new, more refined metric.
