All metrics are generated by making a measurement with our open source software the Green Metrics Tool
The rendering energy shown by WebNRG is the energy consumed by the CPU.
In order to get the energy of the page rendering we utilize an aptly configured dedicated on-premise server that can sample the CPU energy at a 2 ms interval.
This data comes from RAPL.
To isolate the rendering cost from startup of the browser and also from network latency we first download the website completely through a reverse proxy, caching everything local to the machine. Then we destroy the browser context and reload the page from the reverse proxy effectively only measuring the parsing and rendering of the webpage.
To make the measurements even more reliable we tightly accuracy control our measurement machines. See more on this below.
To generate the categories from A+ to F we use the idle power consumption of our measurement machine as capture by the Idle phase in the Green Metrics Tool. We then take the lowest energy consuming page we could find (motherfuckingwebsite.com) as A+ and then linearly increment steps by 25% increase in rendering power. Everything above 2.5*Idle Power is considered F.
The decision for 25% was made because on our measurement system we see a 95p STDDEV of 0.2 W. 25% of the idle power consumption equals 2.5x this STDDEV which is a typical way of determining a range for a statistical relevant difference (outlier) between categories / observations.
Network traffic gets captured from a cgroup in which the browser is executed. This isolates traffic from the webpage from network traffic that occures through noise on the host system.
Different to most other tools on the market webNRG captures the network traffic at the adapter level and not on the HTTP protocol level. This means it includes overheads like package retransfers and thus will slightly differ from values that you can for instance get through the Browsers Developer Tools. We opted for this measure as it more reflects the really transferred data over the network with reasonable reproducability thus also rewarding servers that use HTTP/2 or have a better connectivity overall.
The network carbon calculation happens according to our methodology explained here: Network Carbon Methodology. Please note that this approach differs quite a bit from others like the SWD model used by for instance Websitecarbon.com.
For the network transfer (not the carbon) we use the same grading system as the SWD model.
As said the tooling we use is open source and you can install it on your own machines. Measurements are of course only reliable if you have a tightly accuracy controlled infrastructure.
All infos how to Install the Green Metrics Tool, Setup your own infrastructure, Measuring Websites and configure Accuracy controlled machines are explained in the Docs
Head over to our free tier supported Green Metrics Tool Dashboard to discover more measurements that we do ... even on non-websites :)