In 15 years, traffic Internet has been multiplied by approximately 500 from 2002 to 2017. CO emissions2 associated were assessed at 762 million tonnes in 2018. So let's imagine that, like the Nutri-score, we have a tool to measure our environmental footprint when we click on a web page, allowing us to assign a score between A and G to the HTTP request. Suppose also that these requests are archived, year after year, in a public database such as the HTTP archive. By regularly exploring this database, we could follow the evolution of the environmental footprint of HTTP requests.
It is the ambition that pursues EcoIndex, created in 2014 and which refers both to a set of bonnes pratiques to build a website and a software tool that allows to evaluate several factors for a given URL: its absolute environmental efficiency using a score function on a scale of 0 to 100 (the higher the score, the the better); its relative ecological performance using a score ranging from A to G as is known for household or food devices (Nutri-Score); the technical footprint of the page (weight, complexity, etc.); and the associated environmental footprint (greenhouse gases generated, water resources consumed).
Its objective is to help as many people as possible become aware of environmental impact HTTP requests and to propose concrete solutions to reduce it. If this model based on the technical metrics of the page mentioned above is rather simple to understand, it also has its limits.
The environmental footprint of an HTTP request
Let us first try to understand what this model consists of. You should know that estimating the carbon footprint of human activities cannot be done directly: the method used is generally based on a targeted activity model, relating to the area studied. This is the case for EcoIndex, which only concerns 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 "legacy" version of EcoIndex consists of a plug-in to be installed on the browser and works as follows: the user provides a URL to EcoIndex, which transfers it to the server side. This returns to the browser an HTML page containing the responses to the request. The plug-in measures the footprint of the application, in number of elements of the web page (the number of HTML tags, noted sun), in number of requests in the returned page (requests) and finally calculates the number of bytes of the returned HTML page (size) and which have passed through the network.
These values are fed into the EcoIndex algorithm to measure performance and environmental footprint.
The "3-tier model" and its limits
Additional analysis by an expert is essential for a complete and reliable operational assessment of environmental performance. Indeed, EcoIndex does not take into account the impact of the computer making the request or of a user journey. Only an isolated query of any use is analyzed, such as Nutri-score or washing machines.
Similarly, when the request is resolved 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 this server in the classic sense of life cycle analyzes (LCA), nor of the different network equipment that is crossed between the user terminal and the data center.
However, it allows discussing the models and their attributes that would significantly characterize the environmental impact of the web, reduced to the dimension of HTTP requests. Other positive sides of EcoIndex are that the loading, creation and display of the page in the browser is not simulated and that the three parameters sun, requests et size report an architecture that governs the macroscopic operation of a query on the web, so EcoIndex makes sense.
A tool with imperfect calculations
As part of the device, the environmental performance is calculated on the normalized basis of constant values fixed once and for all and hidden in the model without taking into account variations over time – for example from one period to another such as confinement , vacations, etc., or the geographical location of the user.
Moreover, it is not directly the parameters sun, requests et size which are considered but values corresponding to quantiles, i.e. a small number of values which have been determined by retrieving the three parameters on the URLs from a database of URLs which refers, the HTTParchive .
One can wonder about the stability over time of these quantiles: are they the same in 2023 as in 2020, the date of their determination for the historical EcoIndex? A priori, websites are regularly reviewed to adopt, over time, better eco-design practices – there is no reason for the quantiles to be fixed once and for all.
Another minor remark, for certain sites such as those of the major media, which are dynamic, the value of Ecoindex has a good chance of changing from day to day, but probably not too abruptly, going for example from the rating of A to G.
Indeed, a website, even dynamic, always more or less respects the same template made up of modifiable elements (text, images, background, colors). We replace one text with another, one image with another, without fundamentally modifying things… Vis-à-vis this phenomenon, EcoIndex seems robust to us because this “template” does not change.
However, the AG scores correspond to the EcoIndex ranges of 100-81 for A, and 10-0 for G, without anyone really knowing what it is: how were these different limits determined? Are they equivalent to the quantiles for the EcoIndex measures of the HTTParchive? They are close but do not coincide exactly.
Other parameters to introduce
Finally, the historical model does not lend itself, a priori, to the introduction of new attributes other than 3-tiers in the model.
We could, however, consider adding notions of energy mix and propose a new EcoIndex+ indicator which provides scores oriented towards A for the low-carbon energies used on the client side and on the server side and scores around G if the energies involved are carbon-based. . If the HTTP request goes through a 4/5G mobile, we could also aggregate the impact in CO2 of the operator, which would lead to a richer view of the EcoIndex+.
To be more exhaustive in the attributes to be injected into EcoIndex+, it is necessary for the community to agree on these new criteria and then to establish calculation methods capable of processing a large number of attributes using the machine learning.
An indicator that needs to be improved
From the perspective of environmental impact metrics and best practices in the eco-design of websites, EcoIndex is a simple approach that contributes to understanding the issues relating to the place of digital technology in global warming. The indicator is particularly interesting in the logic of improving successive versions of websites.
However, there is still a long way to go to, on the one hand, deepen our knowledge and better understand the relationships between the different high-level models of the 3-tier architecture type and the field analyzes of the life cycle type of a product or equipment. digital.
On the other hand, it would be appropriate to question the initial model by data science approaches, that is to say, to explore the latter, to analyze them to obtain a new finer metric.
Denis Trystram, University professor in computer science, Grenoble Alpes University (UGA); Christophe Cerin, University professor, Sorbonne Paris North University et Laurent Lefevre, Computer Science Researcher, Inria