Category: Applications sobriety

Netflix, along with others like the BBC, has researched, with support from the University of Bristol, the impact of its service. The precise figures and the methodology will be published soon, but it appears that one hour of viewing Netflix is equivalent to 100 gEqCO2.

When the communication was released, several digital players took up this figure, but, in my opinion, not for good reasons. Communicating the impact of video through The Shift Project emerges as a systematic point of debate. As of March 2020, the Shift post had been widely disseminated in the media with a significant evaluation error. This error had been corrected in June 2020 but the damage was already done.

In this context, the IEA carried out a contradictory analysis on the subject. In the end, many studies on the impact of the video came out (IEA, the German Ministry of the Environment, ourselves with our study on the impact of playing a Canal + video). It is always difficult but not impossible to compare the figures (for example, whether or not the manufacturing stage is taken into account, the representativeness of the terminals, the different infrastructures, and optimizations between players, etc.), however, if we take things comparable, all studies have similar orders of magnitude. By taking the correction for the Shift Project error (Ratio 8 resulting from an error between Byte and Bit), the numbers are also close.

Olivier PHILIPPOT

Digital Sobriety Expert
Books author «Green Patterns», «Green IT – Gérer la consommation d’énergie de vos systèmes informatiques», …
Speaker (VOXXED Luxembourg, EGG Berlin, ICT4S Stockholm, …)
Green Code Lab Founder, ecodesign software national association

The internet browser is the most important tool on a mobile device. It is the engine for browsing the internet. No longer just for websites but also now for new types of applications based on web technologies (progressive web app, games, etc.).

For this new edition of our ranking, carried out in 2018 and 2020, we have chosen to compare 16 mobile applications: Brave, DuckDuckGo, Chrome, Ecosia, Edge, Firefox, Firefox Focus, Firefox Nightly (formerly Firefox Preview), Kiwi, Mint, Opera, Opera Mini, Qwant, Samsung, Vivaldi et Yandex.

The objective of these measures is to see how the solutions stand in terms of environmental impact (Carbon) in relation to each other on common user scenarios but also to provide benchmarks on our uses of browsers.

For each of the 16 applications measured on a Galaxy S7 (Android 8) smartphone, the scenarios integrating the launch of the browser, browsing on 7 different websites, periods of inactivity, etc. were carried out through our Greenspector Test Runner, allowing the performance of automated tests.

Learn more about our methodology

Total energy consumption (in mAh)

The average power consumption is 49mAh (as a reminder, the 2020 ranking average was 47mAh or -4.2%).

Here is the evolution from last year.

	2021 Ranking	2020 Ranking	Évolution
Firefox Focus	1	10	9
Vivaldi	2	4	2
DuckDuckGo	3	5	2
Firefox Nighly	4	10	6
Yandex	5	3	-2
Kiwi	6	8	2
Opéra	7	2	-5
Brave	8	7	-1
Ecosia	9	1	-8
Chrome	10	6	-4
Samsung	11	9	-2
Firefox	12	13	1
Edge	13	11	-2
Qwant	14	13	-1
Opera Mini	15	14	-1
Mint	16	12	-4

Firefox Focus is the best solution in terms of energy consumption in our comparison. The version evaluated in 2020 was one of the first versions and it seems that Firefox teams have been working on optimizing the power consumption of their browser since. Ecosia loses its leading position on this indicator and finds itself in the middle of the ranking. On the side of the most energy-hungry browsers, we find Mint and Opera Mini. Note that the most popular browsers: Edge, Firefox, Chrome, and Samsung, are quite poorly classified.

This total energy consumption can be evaluated and analyzed in 2 ways: the energy consumption of pure navigation and the energy consumption related to the functionality of the browser.

Energy consumption of navigation (in mAh)

Navigation is the consumption only associated with viewing the page (no consideration of launching the browser, features, etc.).

Most browsers have a fairly similar power consumption on “pure” navigation. This is mainly due to the use of visualization engines. Most browsers use the Chromium view engine.

Compared to the 2020 ranking, it seems that the Firefox engine has improved. Qwant, using this engine too.

Energy consumption of features (in mAh)

The functionalities include browser states such as idle periods, launching the browser, writing URLs in the navigation bar.

By keeping the same classification as for the total energy, we see that the non-navigation functionalities (writing of URLs, inactivity of the browser, etc.) have a significant impact on total consumption.

Autonomy (hours)

Battery life is the number of hours the user can surf before the battery is completely discharged. The ranking does not change with respect to that of energy, as autonomy is directly related to energy.

We observe that the autonomy can double from 5h to 10h between the most consuming browser (Mint) and the least consuming (Firefox Focus).

Data (Volume of data exchanged) (MB)

Some applications do not manage the cache at all for reasons of data protection and privacy, use proxies that optimize data, have a difference in the implementation of cache management. In addition, if a browser is good, the downside is that a lot more data is potentially loaded in the background. In our methodology, we see it for the New York Times site, which is larger in terms of data.

Here is an example of the measurement iterations on the Amazon site (Amazon.com) that shows the difference in data processing between different browsers.

Memory consumption(RAM) by the browser process (MB)

Memory consumption is important to take into account in a digital service because even the variation in memory consumption does not influence the energy impact, it remains very important to integrate because of the effects of overconsumption on already congested devices. in memory, or older, less powerful, this can create instabilities or applications that cannot operate simultaneously because they compete. In ecological terms, this can of course provides a premature change of device on the user side for a more powerful model to satisfy good user comfort.

The variation goes from 400MB to 1.8GB (approximately half the RAM of the Samsung Galaxy S7).

Let us observe more precisely the behavior of the memory following the sequence:

• Launch browser
• Browser inactivity
• Navigation (Average memory consumption)
• Inactivity following navigation
• System after closing browser

At the launch of browsers, we have a median memory usage of 413MB. Edge consumes a lot more with 834MB.

If we leave the browser inactive, the memory consumption of most browsers remains fairly stable. Which is pretty good and normal. On the other hand, we see that Edge and Ecosia have a strong increase in memory.

Then, with navigation, the memory consumed increases significantly. This is due to the consumption of navigation engines to analyze and store items. The management of tabs will also play a role. If the browser offloads the memory for the non-active tabs, then the consumption will be lower.

We can note that Firefox Focus, Mint, Duck Duck go, Opera Mini and Qwant overall consume little memory.

When the browser is closed, almost all browsers are no longer in memory. Firefox remains however with 1, GB as well as Chrome and Mint with around 100MB. Probably a bug but it is annoying because elements still occupy the memory and processing operations can also exist: processing operations are confirmed on Firefox and Mint with the rate of CPU consumed by the browser process which remains high.

We can also look at the memory impact of consulting Wikipedia (the basic consumption of the browser is subtracted here).

We understand the difference in memory management between browsers and the potential entropy on heavier sites.

Performance

We measured the time it took to write the URL in the address bar.

This difference in performance can be explained by several factors: network exchanges during entry (auto-completions), processing during entry, search based on known addresses, etc. In the end, for the user, the time to access the site will be longer or shorter. For example on the Wikipedia URL entry on Duck Duck Go a lot of network traffic and CPU processing (peak at 22% CPU).

Unlike the faster Edge which has lower processing in terms of CPU.

By the way, we could have an optimization of all the browsers by limiting its treatments (for example by grouping and spacing the treatments).

Environmental impact

The environmental impact is calculated according to the Greenspector emission factors taking into account the energy consumed and the wear of the battery (impact on manufacturing). The impact of the network and the data center is taken into account with the internet intensity.

This impact is reduced to the consultation of a page.

Firefox Focus by its low consumption is first. Samsung, which has average power consumption, is in second place thanks to good data management.

The most impactful browsers (Ecosia, Edge, Mint and Opera Mini) have high power consumption and poor data management..

Olivier PHILIPPOT

Digital Sobriety Expert
Books author «Green Patterns», «Green IT – Gérer la consommation d’énergie de vos systèmes informatiques», …
Speaker (VOXXED Luxembourg, EGG Berlin, ICT4S Stockholm, …)
Green Code Lab Founder, ecodesign software national association

Introduction

This study was carried out by Greenspector, a company specializing in the impact of digital technology, and EVEA, specializing in analyzes of the environmental impact of products. This study is based on:

• a calculation methodology initiated as part of the CONVINcE project on the energy consumption of video. A project involving several players such as Orange, Sony …
• energy and resource consumption measurements carried out on the myCANAL application to support hypotheses
• collection of data on the use and infrastructure of Canal +
• a bibliographic study to identify and correlate emission factors and energy consumption

Study summary :
– Methodology
– Results : consumption, overall impact and projection
– Areas for improvement
– Conclusion – points to remember
– Chief Digital Officer – Canal+ testimony

Definition of visualization

Visualization quality influences end-to-end power consumption. We have taken the following 3 main definitions:

• SD (Simple Definition)
• HD (High Definition)
• UHD / 4K (Ultra High Definition)

The definition depends on 3 factors:

• compatible program availability
• material compatibility in terms of quality
• the quality of the connection (the quality of myCANAL videos is adapted according to the connection speed).

Greenspector’s exploratory measurements on myCanal show the impact of SD, HD, or UHD video consumption on smartphones, and then on laptops:

Bibliographic sources were used for the consumption of other platforms such as Consoles (Source 1), TV (Sources 1 and 2), Set-Top-Box (Sources 1 and 2). The consumption of Canal + decoders supplied by Canal + was also used.

Depending on the different equipment configurations, we obtain energy consumption which varies from 1.3 to 108 Wh / h, i.e. a ratio of 1 to 80:

Means of connection to the network

To access the services, the infrastructure considered takes into account: the user’s equipment (Box among others), access to networks (GSM, Fiber, etc.), and the heart of the IP network.

We used calculation methods that are widespread in the scientific literature and standardized by ETSI. The principle is to take an “energy / typical use” ratio in Wh/Go. Although the network infrastructure has a fairly fixed consumption and does not depend on the use, this calculation method makes it possible to assign a global impact to a use (here one hour of video). In addition, it allows to study the improvement of networks in terms of efficiency. Beyond the search for improved energy efficiency, it also makes it possible to assess pressure on the network and to take into account the improvement in the capacity of the infrastructure.

As explained in the CONVINcE report:

“As we are looking for an order of magnitude in energy saving, we suppose that decreasing by 30% the traffic volume in the core network induces a decrease of same ratio in network dimensioning and consequently a decrease of 30% in energy consumption in the core IP network.”

We have studied the literature and listed metrics ranging from 1.3Wh / Go (figure for FTTH fiber) to 600Wh/Go. These factors fluctuate depending on the infrastructure assessment date, method (Top Down or Bottom Up) and technology. It is clear, however, that the efficiency improves over the years.

We have taken:

• 13 Wh / Go for the core network (source CONVINcE)
• 30 to 40 Wh / Go for the fixed network
• 150 Wh / Go for the GSM network (source CONVINcE)

For TNT and satellite, there are few data. However, we based ourselves on a BBC study for TNT.

Content Delivery Network

Content Delivery Networks (CDNs) are servers used to limit the load on “Top of Head” servers (serving videos) and to provide a flow as close as possible to the user. Canal+ uses CDN providers on the market but also uses its own infrastructures.

The methodology for estimating the energy intensity per GB is to bring the estimated consumption of the data center down to throughput during peak periods. Canal+ indeed knows the number of physical servers, their types as well as their speeds. As the servers are hosted by a host, certain assumptions have been made to estimate the actual consumption: among other things, an assumption of PUE (Power Usage Effectiveness) of 2, consumption per server of 250W, and a server load of 50%. The consumption of routers and storage is considered negligible compared to the consumption of servers. (Sources 1, 2 , and 3)

The energy intensity obtained is 0.13 Wh / Go, with a different value between VOD and Live. Note that this value is calculated in the event of a peak (football match for example). It can be lower (use of the total capacity of the data center) or higher (during periods of low traffic).

For lack of data, the hypothesis of the same energy intensity (Wh / Go) was taken for CDNs outside Canal + (50%). For comparison and verification, the studies listed display values from 0.04 to 1 Wh / GB. These differences can be explained by various factors:

– Improving the efficiency of data centers, old studies, therefore, lead to higher figures,

– Top-Down approaches that have a higher estimate than Bottom-Up studies (like this estimate on CDNs).

AWS application servers Excluding video: using Mycanal (catalog presentation, authentication, etc.) and watching videos involve requesting services hosted on servers (rights management, etc.). The APIs used are hosted on AWS instances.

Canal+ rents AWS instances and knows the number of VMs. A part of these VMs is purely dedicated to providing ancillary services to video.

It is difficult to know the power consumption of AWS instances because no communication or information is provided by Amazon. According to an expert, we took a value of 20W (taking into account a PUE of 2) (Source: Interview of experts on virtualization from the company Easyvirt).

We have assumed a uniform distribution of consumption concerning the number of hours visualized and we obtain a value of 0.14 Wh / h.

Video encoding servers

The “Headend” servers allow you to format videos, “package” them to the user’s format … For the OTT, up to 150 packages may be available for a single video. Here is the workflow for the OTT.

The most consuming parts are video encoding and decoding. The workflow is distributed over specific servers (for encoding) and classic servers (for formatting and packaging).

We took the figures from the CONVINcE project (considering Harmony servers identical to those used by Canal +) to estimate the energy, as well as the Canal + server data:

• For IPTV: 0.09 Wh for one hour of video
• For the OTT: 0.30 Wh for one hour of video

This difference is explained by the fact that OTT is encoded in several formats, unlike IPTV which is encoded in a high definition format.

Results – global projection

The supply of renewable energy to AWS infrastructures and Canal + CDNs hosted by Interxion has not been taken into account.

In one hour, on average, here are the flows that pass through the network:

500 000 hours of video

900 TB of data

3.6 GB / hr average throughput

The number of viewing hours for Canal+ subscribers has been broken down according to the parameters listed above.

The unit data obtained previously are then projected on these uses to obtain the overall consumption. By taking these results, we can obtain the usage intensities of each part (Terminal, Network, Server) to check the consistency of the overall consumption.

Intensity analysis

The usage intensities obtained can be compared to the Shift Project and IEA studies. This intensity is calculated by taking the distribution of devices, the use of Canal +, and keeping only the IP transfer part (without Satellite or TNT).

It is clear (and shared by the IEA) that the Shift Project study overestimates grid consumption. For servers, we have a fairly low value but are quite confident considering that it is based on data from a known infrastructure (with the uncertainty from Amazon servers and third-party CDNs). The estimate of the terminal part of the Shift Project seems to us to have been underestimated (shared by IEA). It seems that the IEA also underestimated many parameters and seems not to have taken into account the influence of the definition on the consumption of the terminal, certain elements such as boxes, and the consumption of new TVs that consume more. The CONVINcE study and the Sauber/Koomey analysis also validate the consistency of our study.

Impact of playing an hour of video

Across the Canal+ fleet, the average end-to-end consumption is 214 Wh per hour of video. For comparison, the IEA study announces consumption between 120 and 240Wh per hour. The breakdown is as follows:

The impact in average CO2 equivalent by type of access is as follows (with the assumptions specified in the paragraph CO2 emission factors where the manufacture of Satellite, DTT, network and server hosting infrastructures has not been taken. into account)

By taking the type of connection:

The impact ranges from 20 to 66 g CO2 eq, it depends on several parameters:

• the more or less energy consumption of the device (for example Smartphone vs TV),
• embodied energy and material life,
• the means of access to the network (for example GSM access has a greater impact than wired access),
• the viewing quality which will influence above all the share of the network (depending on a cost per Wh / Go),
• the number of materials to access the service.

By taking real uses (see the end of the document), the average is 28g CO2eq. If we take only the IP part, the average value is 37g CO2eq.

For comparison, the IEA study (2020) announces 8g CO2eq for France (for the use phase only). Another study for the US (2014) announces 360g CO2eq on the use phase (and 420g CO2eq with manufacturing). If we take a US energy mix, we get 202g which brings us closer to this study. The lower value can be explained by the differences in assumptions, in particular on the energy intensity of the network. For example, if we take for our study a TV with 4K reception, IEA announces 20g CO2eq while we rather estimate 14g (37g with the manufacturing phase).

If we look at the usage / manufacture ratio, in some cases the use has more impact (UHD in particular), while in others it is the manufacturing (IP TV or PC in HD for example).

In this same study, the estimate of the purchase of a DVD is 400g EqCO2, which therefore leads to the impact of watching a 2-hour video. 5 times less than that of buying a DVD in France and equivalent in the US.

Consumption and overall impact

The result of the total energy consumption of viewing Canal+ videos over IP is 900 GWh per year with a greenhouse gas impact of 159,000 tonnes CO2eq.

For comparison, annual French consumption is 473 TWh per year (Source RTE) and France’s 2017 carbon footprint (national emissions + imports) is 749 Mt CO2 eq (source: Haut Conseil pour le Climat – 2019 report). The consumption of the Canal + fleet by IP is therefore 0.18% of energy consumption and 0.016% of the French carbon footprint.

The end-to-end distribution is as follows:

Most of the consumption is on the user’s premises (terminal and part of the network access). Indeed, it is necessary to have the equipment (TV, Box, Smartphone …) which is not shared like the servers.

To validate the consistency of this projection, we took the consumption of CDNs specific to CANAL as well as to suppliers. We have a consumption of Canal + CDNs of 2.6 GWh (and estimated at 7 GWh for suppliers) excluding video head-ends which are not at Interxion. We get 12 GWh with the projection, which validates the model.

If we look at the distribution of the impact in greenhouse gases, we have the following distribution:

A large part of the impact comes from the manufacture of user terminals. On the network part, the manufacture of Boxes (FAI, Satellite …) also has a significant impact.

Areas for improvement to limit the impact of services

Strengthen bitrate downsizing

To date, except in Africa, or on a cellular network (so as not to empty the customer’s subscription), the quality of the video is adequate . On PC, quality adaptation is possible but access is not necessarily easy.

Making the possibility of reducing the size more accessible by different means (improvement of the interface, communication, etc.) would make it possible to redirect some of the users to a lower but sufficient quality.

Taking into account the increase in video quality and that of network performance, quality caping by users of Canal + services should represent in 3 years a gain of 20% in average speed compared to that generated by uncapped use.

In this axis, several elements have not been included in the gain projection but are possible to help the user in his impact. For example, the Greenspector mobile measurements show that it is better (under favorable conditions such as wifi) to use downloading rather than streaming. Indeed, here is the comparison for a 45mn video (with 5mn download):

This is partly explained by the fact that when viewing the downloaded video, the radio cell is not used while it is much more during streaming.

Improve the interface and the software layer

Part of the impact of viewing an hour of a video comes from the interface. Indeed, viewing the catalog, managing subscriptions (via APIs), processing playlists … are necessary functions. Greenspector measurements showed that this could represent 10% of terminal consumption:

Improving the interface on all platforms (smartphone, PC, etc.) would therefore make it possible to obtain significant gains. It would also allow, beyond reducing the impact, to limit the exclusion of certain people as well as the obsolescence of platforms.

Among the actions identified:

• Switch to the dark mode
• Reduce the overall impact of the software layer
• Improve UX
• Limit the integration of third-party libraries

Results

Here is the projection of the various energy gains:

The improvements compensate for the increase in energy consumption and even allow a gain of 14% compared to the current situation. For the greenhouse gas impact, we obtain a gain of 8%.

There are many ways to reduce the carbon impact of streaming activities for Canal + by 30%, and the actions put in place will make it possible to make a gain of 26% on the environmental footprint of services and in particular -31% for the consumption in OTT only.

Pierre-Emmanuel Ferrand, Chief Digital Officer at CANAL+

Testimony

– How does this assessment and improvement process fit into the Group’s strategy?

CANAL+ is anchored in its time. We have decided to become more involved in a major issue of our time, dear to our subscribers and users of myCANAL: the protection of our environment. CANAL+ was very early on as a pioneer in eco-responsible approaches. first. In addition to the rental model of decoders, which is very virtuous from the point of view of the circular economy, Canal+ organized in 1988 the recovery of old decoders to ensure their return to service or their recycling. Moreover, as a publisher, We have produced eco-responsible productions including series that have been exemplary on this subject, Baron Noir, L’Effondrement, and more recently OVNI (S).

-What main lessons/contributions do we draw from this work (on the design of services)?

This work allows us first of all to identify the technical projects to prioritize to minimize the carbon footprint of myCANAL video consumption as much as possible. At the same time, the lessons guide us on the awareness messages to relay to our users, throughout our future developments. This commitment to cooperation between our technical developments and our users is the key to consumption that has less impact on the environment. 

– What role do users play?

Protecting the environment is a priority for them. Their role is central to our approach. Our challenge is to offer them a platform offering the best content and the best video quality adapted to their equipment and their reception capacities to reduce the effects on the environment. By raising their awareness and supporting them in responsible actions, they too will be able to get involved, as they do in other areas for a more eco-responsible society.

Olivier PHILIPPOT

Digital Sobriety Expert
Books author «Green Patterns», «Green IT – Gérer la consommation d’énergie de vos systèmes informatiques», …
Speaker (VOXXED Luxembourg, EGG Berlin, ICT4S Stockholm, …)
Green Code Lab Founder, ecodesign software national association

Public organizations have particular challenges when it comes to digital eco-responsibility. Like private organizations, they respond to a global challenge to limit the environmental impacts of digital services, which are growing as much as public services are modernizing and rapidly becoming computerized.

This subject, like accessibility to the greatest number of people becomes a subject of exemplary nature for public services which, we hope, will have a vocation to have a capacity to train private organizations and more generally society at large. Eco-responsibility has appeared several times in the political sphere since we see elected officials who seized on the subject at the same time as political dissident movements integrated this dimension of eco-responsible digital. As a sign of this enhancement and of the new speeches / political programs, we now have an elected representative in Responsible Digital for the city of Nantes (France).

Beyond the environmental issue, digital sobriety also helps to reduce the digital divide, allowing equal opportunities for online content and services since a more frugal service will also be more accessible to citizens with limited or limited connection. an old-tech, low-tech or cluttered computer / mobile.

An acceleration is underway for the consideration and will become even more visible since article 55 of French Law n ° 2020-105 of February 10, 2020 relating to the fight against waste and the circular economy »Specifies that organizations must promote in the markets software whose energy consumption is limited during the use phase.

A digital eco-responsibility approach is a quality approach that comes at a cost

But how to integrate these requirements in the clauses of the contracts, how to manage them and how to verify them?

We can read some requirements:

Often awkward or superficial, digital eco-responsibility requirements must be more precise and be oriented towards results objectives rather than means requirements.

What type of requirement?

Results requirements have several advantages over good practices (means requirements) because they absolutely validate that the result is good or not good without an expert debate, without a debate on the applicability of a rule, partial application or not, correct implementation or not.

They are above all not very dependent on technological developments and have the advantage of remaining on intermediate indicators that are easy to collect and are easily verifiable and measurable.

However, they must be accompanied by a repository of best practices, in the appendix, to avoid putting in place corrective actions to optimize after the fact. It is also a good way to involve the design and development teams in the process. These attached documents should not be too heavy and focus on generic standards, not techno or language oriented.

They have 2 complexities: they must be calibrated to set thresholds and they do not transfer all responsibility for the result to the person who “manufactures” (MOE).

What metrics to measure to cover the major impacts of a digital service?

A high level indicator comparable to other areas is the carbon impact expressed in geqCO2. The advantage of this indicator is that it is universal and can be shared by everyone in the company, with its customers, comparable with its competitors, in its ecosystem, with any service / product.

The carbon impact is not a measurable indicator but “projectable” on the basis of flow or intermediate indicators. These include energy consumption, network data consumption, display performance or the number of requests to servers internal or external to the organization.

In any case, the Carbon projection will be all the more reliable if we start from real measurements on real devices than from estimation or measurement on an emulator. This projection can be improved by knowing the routes or the type of material on which the service or the content is used.

How to set thresholds for a result requirement?

Setting thresholds first requires specifying the conditions under which measurements such as:

Setting thresholds also requires benchmarking one’s domain or existing sound in order to understand the values. By fixing such a value, we answer the 2 questions, am I in values of good qualities and are they attainable?

Means objectives can be added as appendices to the results requirements. A repository of good practices will have a positive impact and allows the MOE teams to get it right the first time without additional correction costs, but be careful, these repositories of good practices or Green Patterns must be instantiated by technology or by languages and therefore evolve over time. pace of technological change or remain very generic.

Validate compliance with a requirement in a project

Once the requirements have been set, they must then be managed during the project via intermediate points which make it possible to probe a partial result and thus avoid the tunnel effect. All these intermediate reviews make it possible to develop automatic scripts which will allow measurement to be carried out in a reliable and comparable manner and to be able to consolidate the test heritage over time. These views also make it possible to adjust if necessary and discuss over-functional or over-content or additional on-board intelligence costs between MOA and MOA.

It is in this validation phase that the result requirement finds its full meaning since we do not need to audit all screens / pages on a repository of good practices but just automated scripts during the project which are replayed on the finished product.

Thierry Leboucq

The stay-at-home context has mechanically increased the mobile applications use of the social network type in order to keep people connected. Like the professional use of videoconferencing tools, these uses have brought additional pressure on the network and on the servers of these solutions.

What are our activities impact on social networks? What are the most / least impactful solutions for the environment, network congestion and the autonomy of our smartphones?

For this study, we’ve chosen to measure the news feed of the 10 most popular social media applications: Facebook, Instagram, LinkedIn, Pinterest, Reddit, Snapchat, TikTok, Twitch, Twitter and Youtube. Although these applications are different in terms of functionality, we have chosen to compare them in terms of carbon impact, energy consumption and data exchanged.

For each of its applications, measured on an S7 smartphone (Android 8), the user scenario lasting 1 minute was carried out through our GREENSPECTOR Test Runner, allowing manual tests to be carried out. For each of its applications, the user scenario corresponds to a scrolling of the contents of the news feed of an active account.

Each data measurement is the average of 3 homogeneous measurements (with a small standard deviation). The consumption measured on the given smartphone according to a wifi type network can be different on a laptop PC with a wired network for example.

To make Carbon projections to assess the impacts of infrastructure (data center, network), we relied on the OneByte methodology based on real data measured by the volume of data exchanged. This evaluation methodology takes into account the consumption of resources and energy in use for the equipment requested. As it is a very macroscopic approach, it is subject to uncertainty and could be refined to adapt a context, a given tool. For the Carbon projection, we have assumed a projection of 50% via a wifi network and 50% via a mobile network.

To make the carbon projection of the mobile, we measure both the energy consumption linked to the use case based on real measurement on real device and to integrate the share of material impact, we rely on the rated theoretical wear and tear generated by the use case on the battery, the first wearing part of a smartphone. 500 cycles of full charges and discharges therefore cause in our model a change of smartphone.

Projected data measured in Carbon impact (gEqCO2)

Youtube (0.66 gEqCO2) is first in the ranking, followed closely by Facebook (0.73 gEqCO2) and LinkedIn (0.75 gEqCO2). This is easily explained, since the only videos launching during the news feed for Youtube are thumbnails and this, after 2 seconds. It should be noted that in our test, the scanning of the news feed was slow enough to launch videos according to this principle of tempo.

On the Environmental Impact side of applications, the social network whose news feed is the most impacting is Tik Tok. Unsurprisingly, this social network is based exclusively on watching videos and preloading videos from the news feed at startup. The Shift Project also presents streaming platforms such as Netflix, Youtube and Tik Tok as being responsible for 80% of digital electricity consumption. We had already noted this significant impact, in particular when the application was launched in 2019.

Only 4 applications (Tik Tok, Reddit, Pinterest and Snapchat) are above the average carbon impact (2.1 qEqCO2) observed for this comparison of the news feed. Moreover, the Tik Tok news feed has a carbon impact of 7.4 times greater than that of Youtube.

What if we were to project all of this at the user level?

According to the Global Web Index 2019, we spend on average 2 hours and 22 minutes on social networks. If we project the average carbon impact of the 10 applications measured (2.10 gEqCO2) over 60 seconds at the average time spent per user, we obtain for a user/day: 280.5 gEqCO2. Or the equivalent of 2.50 km traveled in a vehicle. This also corresponds to 102 kgEqCO2 per user per year, the equivalent of 914 km traveled by medium vehicle in France. This is equivalent to 1.5% of the carbon impact of a French person (7 Tons).

And globally?

The 2019 figures of LyfeMarketing and Emarsys announce 3.2 billion social network users (42% of the world population) of which 91% access social networks via a mobile device. 80% of the time spent (2 hours and 22 minutes) on social networks is spent on a mobile device. If we project our carbon/user impact to these data, we obtain: 262 million Tons EqCO2 per year for the 3.2 billion users on mobile, the equivalent of 56% of EqCO2 impacts in France.

Energy consumption measurement (mAh)

In terms of energy consumption, the bad students are the news feeds of the Snapchat, Tik Tok and Pinterest applications. The good energy students are Youtube, LinkedIn, and Reddit. The Snapchat news feed consumes 1.6 times more energy here than that of Youtube.

The average established for energy consumption is 10.6 mAh, only 4 applications are above.

If we assume that the application runs continuously on the smartphone, then we can project the remaining battery life time (graph below). We can observe that with Snapchat running, the battery lasts 3.45 hours. On the Youtube side, the battery autonomy lasts 5.46 hours, i.e a ratio of 1.5 (or a difference of 2 hours) between the best and the least good application of this ranking. The average is 4.8 hours for all of these applications. The reference measurement of the test smartphone is 1.32 mAh, its battery capacity of 3000 mAh, we can estimate its autonomy at 18 hours. The use of social networking applications therefore greatly impacts your battery life.

Measurement of data exchanged (MB)

In terms of data exchanged, the bad students are the news feeds of the Tik Tok, Reddit and Pinterest applications. The good students on the data exchanged side are Youtube, Facebook and LinkedIn. Tik Tok consumes 9 times more data than the Youtube application.

The average established for the data exchanged is 19.2 MB for this use. Beware of your data plans! Projection in 1 month, you will have consumed 74 GB!

Taking into account the real average time spent by social network according to the Visionary Marketing blog: if you only use Tik Tok in social network (up to 52 minutes per projected day), you will consume nearly 71 GB per month, while Instagram (up to 53 minutes a day) will consume 25.6 GB! Are you more connected to Facebook? This will make you consume almost 10 GB (up to 58 minutes per day) per month.

For those who like numbers

Kimberley DERUDDER

Kimberley DERUDDER has been digital marketing manager at Greenspector for more than 5 years. Kimberley graduated with a master’s degree in Marketing – Communication and specialized in Inbound Marketing after her first two years at Greenspector. Today in charge of the animation of the marketing, social media and lead generation strategy, she also takes care of app comparisons and battles.

Article updated with new measures on StarLeaf, Rainbow and Circuit by Unify on May 19, 2020.
Article updated with new measures on Hangouts and Google Meet on May 4, 2020.
Article updated with new measures on Tixeo and Infomaniak Meet on April 23, 2020.
Article updated with new measures on JITSI and Teams on April 15, 2020.

The current “stay at home” context mechanically increases the use of online collaboration tools, in particular videoconferencing tools. This leads to a pressure on the network and more particularly to a significant load on the servers of each solution. It is therefore important from an efficiency point of view but also an environmental impact one to choose the simplest and most efficient solution.

For this study, we compare 14 videoconferencing applications: Whereby, Webex (by CISCO), Skype, Zoom, JITSI, Microsoft Teams, Tixeo, Infomaniak Meet, Hangouts, GoToMeeting, Google Meet, StarLeaf, Rainbow and Circuit by Unify.

For each of these applications, measured on an Samsung S7 smartphone (Android 8), the following three scenarios were carried out through our GREENSPECTOR Test Runner, allowing manual tests to be carried out over a period of 1 minute:

• Audio conference only in one-to-one
• One-to-one audio and video conference (camera activated on each side)
• One-to-one audio conference and screen sharing

Each measurement is the average of 3 homogeneous measurements (with a small standard deviation). The consumption measured on the given smartphone according to a Wi-Fi type network can be different on a laptop PC with a wired network for example.

Measurement of energy consumption (mAh)

The StarLeaf application consumes the most of all three scenarios. This is due to the fact that the consumption in audio only mode, audio + screen sharing or audio + video is the same. This is a special case of our study. The GoToMeeting application is the least energy-consuming, closely followed by Hangouts, Zoom and Webex.

The energy consumption of all these applications is on average 2.1 times higher when adding video to audio and only slightly higher when adding screen sharing to audio (+14%). It’s not a surprise, avoid sharing with a camera to consume less energy on your devices during your video conferences and save your autonomy and the lifespan of your battery!

Measurement of data exchanged (MB)

GoToMeeting and Webex are the two applications that consume the least data. JITSI and Infomaniak Meet are the ones that consumes the most. Overall and without any surprise we note that the audio scenario is the one that consumes the least data. While the scenario activating both audio + video stream is the most consuming one.

The data consumption of all of these applications is 13 times higher when adding video to audio and almost doubled when adding screen sharing to audio (+77%). It’s not a surprise, avoid sharing video and limit your screen sharing to consume less data on networks in your video conferences!

It should be noted that these significant differences are mainly due to the significant audio-video consumption of the JITSI application with 35 MB transferred in 1 minute compared to 1.13 MB for GoToMeeting! JITSI‘s optimized mode does not reduce this data impact (33.4 MB / minute). Infomaniak Meet based on the JITSI engine meets the same volume anomaly on average and mainly on the audio + video part without improvement with the optimized mode.

Projection of the measured carbon impact data (gEqCO2)

A conference on mobile is 3 times more impactful for the environment when we add video to audio.

The carbon impact projection of all of these applications is on average 3 times greater when adding video to audio and higher when adding screen sharing to audio (+35%). It’s no surprise, avoid sharing video and limit your screen sharing to less impact infrastructure (network, datacenter) and on your devices in the context of video conferences!

Without any surprise, a large part of the carbon impacts are located on the network part (63%) but the impacts part on the device (28%) should not be overlooked!

The JITSI and Infomaniak Meet apps even averages 5.8 times more impact when adding video to audio and 40% more when adding screen sharing.

In the Skype vs Teams battle at Microsoft, the overall results are very close (6%). The Carbon impact is lower for Skype with a lower data impact but higher energy consumption on the mobile than Teams.

Which applications optimize energy and data consumption?

Only Whereby and Webex could be tested on optimization features for the mobile version.

The Whereby application with its “Mobile mode” setting which limits the resolution of the stream as well as other resource consumption optimizations. And Webex thanks to its configuration allowing to deactivate High-definition playback of videos only. Whereby saves 21% on energy consumption on the audio side, 15% on the audio and video side and 3% on the screen sharing side. These gains are nevertheless low considering the results obtained by the application that consumes the most audio and video mode from our panel.

On the Webex side, the gains are barely visible as limited only to the video part. It’s barely 5% gain in the Audio and Video scenario for the energy and carbon impact. The exchanged data is even slightly higher in audio and video mode.

The optimized JITSI application does not improve the data or energy impact. Power consumption is even higher in screen sharing mode in this optimized mode. However, there is a reduction in the data loaded in this optimized but minor mode (11% on average). Same findings for Infomaniak Meet.

The optimization of the TIXEO parameters makes it possible to considerably reduce the impact in audio + video mode and allows it to be classified as the 2nd least impacting application in the panel.

As for the Zoom application, a notification of battery overconsumption appeared several times during the measurements. Although we can optimize the video quality on PC, it does not seem possible to configure this on mobile. Without this optimization, the application becomes the most energy-consuming application in audio + video mode.

GREENSPECTOR advices:

• Promote audio only during your conferences: the video stream will tend to consume much more especially when sharing video via camera.
• We recommend that publishers provide optimization options to the user and make them as accessible as possible even by default.
• Prefer videoconferences over travel by car!
• Comparison for 2 people who talk to each other for 3 hours in video + video (1 gEqCO2 on average per minute) when one of the two made 20 kms (112 g eqCO2 / km in France) round trip for a face-to-face face,
• In videoconference: 180 * 1 * 2 = 360 gEqCO2.
• By car: 112 * 20 = 2.4 kg EqCO2. Or 6.7 times more for the physical one-to-one.

Thierry Leboucq