Re: How many emissions in a GB?

This is great, thanks. Some very useful considerations and links I will
definitely be following up. Your points all make sense to me, and you may
have noted I did not suggest any particular formulae, practices or
standards, although I do mention LCAs.

As an evaluation professional, I know that evaluation is always contextual,
so the engineering team benchmarking its code emissions to keep within
approximate budgets or achieve and confirm reductions, may or may not adopt
or require an LCA in order to achieve its goals.  A company with ESG
reporting requirements may indeed need some form of LCA. An academic
exploring outliers, may dispense altogether with industry standards to
achieve much narrower but more precise granularity.


From this perspective I think there are two aspects to this.

On the one hand, the kind of incredibly helpful and comprehensive averages
you have compiled. These are fantastic, rigorous benchmarks that can be
adopted for projections, estimates, emission budgeting and trend tracking.
They are an example of what I suggest as a starting point: pick a
scientifically backed metric, and build on top of it.

On the other hand, these are averages, which by their nature tend toward
the middle of a distribution. Your averages are particularly powerful
because of their range of scenarios, and the transparent and numerous
indicators applied for each scenario. But they are not precise predictors
of actual emissions. My post identifies the factors that can cause a
measurement to deviate from the averages. It becomes impractical if an
average calculation has to include over 100 scenarios, each with over 700
indicators.

But you would need something like that to span the range of interactions
between a user's device, the device model, the OS including the eye
watering range of Linux distros, each with a different energy consumption
profile, the amount of processes, tabs, applications, a user has on the go,
the electricity grid at the server and at the client zones, the amount of
web trackers in operation, the presence or absence of specific malware, the
api architecture and protocol, the kind of peripherals involved, a user's
reading speed, the stickiness of the web content, etc., etc., etc.

I loved Gerry McGovern's piece, linked in my post, on the emissions
generated by writing a web article. When you take account of the time your
readers will spend on their devices reading your blog post, the emissions
will dramatically vary with the author's popularity, the length of a
particular post, its unique virality, the language in which it's written (I
have friends blogging in Greenlandic) and the regions and devices where it
is primarily consumed.

So my suggested approach is, begin with evidence based averages, like the
ones you have shared, which sound to me like best-in-class examples. And
then, iteratively, replace them if you have the capacity,  with actual
measurements for your own unique cases. Some of these will conform to the
averages, while others will be outliers that would be unreasonable to adopt
as general benchmarks, but which are in fact the closest to your own
product's realities.

Hope this makes sense, and I will absolutely be diving into the material
you shared. I would love any further resources and documentation you might
be able to share on your metrics, scenarios and indicators.

Collegially,

On Mon, 22 Aug 2022, 02:43 Frédéric Bordage, <info@greenit.fr> wrote:

> Hello Ismael,
>
> Thanks for this try.
>
> There are several important points to notice.
>
> 1. No linearity
>
> Network.
> There's no linearity of environmental costs for fixed lines (DSL, fiber)
> and HDD. That means that we should better not divide a number of GB
> exchanged by the environmental cost of the infrastructure. This is
> nonsense.
> One the other hand, environmental impacts of 4G / 5G are much more linear.
>
> Storage.
> Same situation for storage. We should better not divide the environmental
> cost of producing and using an hardrive disk (HDD) by its capacity of
> storage. This is non linear.
> One the other hand, environmental impacts of SSD are much more linear.
>
>
> 2. LCA methodology
>
> From a methodological perspective, one should better use the Life Cycle
> Assessment (LCA approach) which is based on standards (ISO 14040 and ISO
> 14044) and is commonly used in most of the world to assess environmental
> impacts. In Europe, where I'm based, you must use this methodology to
> assess the environmental impact of digital stuff. See
> https://ec.europa.eu/environment/eussd/smgp/ef_methods.htm
>
>
> 3. NegaOctet and EcoIndex
> As part of the NegaOctet.org and EcoIndex (see https://github.com/cnumr/)
> projects, we already calculated an average environmental cost for a web
> page.  The first approach (NegaOctet) is based on an LCA modeling peer
> reviewed by a French public research body. The second project is based on
> another LCA of one of the top 10 French website.
>
> Environmental impacts already calculated:
> Write, send and read an email
> Watch 1 hour of streaming video
> Download or upload
> Store in the cloud
> Set up a webconference
> Set up a audioconference
> Search for an information
>
> For each of these functional units, we have several scenarios based on
> different parameters. And for each scenario, we provide 29 environmental
> impacts - Global Warming Potential, Ionising radiations, Abiotic resources
> depletion, Water Usage, etc. - based on international and european
> standards (ISO 14040/44, PEF, etc.).
>
> If it's of interest for the group, I would ask my partners if they allow
> me to provide some web environmental impact factors to this working group.
>
> Best,
> Fred
> +33 6 16 95 96 01
> GreenIT,.fr founder
> We provide data about the digital world's environmental impacts.
>
>
> Le lun. 22 août 2022 à 01:13, Ismael Velasco <ismaelv.dev@gmail.com> a
> écrit :
>
>> I thought this might be of interest to the community, in terms of the
>> need to choose metrics for measuring the carbon impact of the applications
>> we design.
>>
>> https://ismaelvelasco.dev/emissions-in-1gb
>>
>> I've written a blog that goes into the range of factors involved in
>> determining the CO2 emissions of data transmissions.  I've all of these
>> referenced in various articles, but haven't come across one that references
>> them all in the same place, with tools and strategies for choosing how to
>> evaluate and monitor emissions from data.
>>
>> TLDR: There is no straightforward metric available (possible?), and the
>> emissions of 1GB will vary by hardware, software, use case and grid
>> intensity. More particularly the emissions will vary by source, device,
>> model, signal type, transfer protocol, active software, use case and grid
>> energy source at a particular moment.
>>
>> I give a brief intro to each in my article, and recommend the focus be on
>> improvement over exactitude, emission reduction over precision tracking,
>> iterating over time to improve and refine metrics.
>>
>> This would be relevant when it comes to issuing guidelines, or
>> integrating emissions tracking into browsers, in dev tools or more
>> prominently. Likewise when it comes to green web certification projects.
>>
>> Appreciate any feedback!
>>
>