A CLI tool to estimate the carbon emissions produced by AWS EC2 usage.
This tool needs an AWS Cost and Usage Report as input. These reports are delivered automatically into an S3 bucket. Usually they cover usage of (up to) one calendar month. Time resolution (hourly, daily, monthly) should not make a difference, both hourly and daily have been confirmed to work fine.
One such report is required to be accessible, e. g. downloaded to the local hard drive. The file is expected to be a gzip compressed comma-separated value (CSV) file.
If you don't have Cost and Usage Reports configured, please check the AWS documtation regarding setting this up.
TODO. Short version: clone the repo and build the binary using go build. Alternatively, download binary from release.
The CLI is invoked as
cloud-carbon analyse PATHwhere PATH must be replaced with the path to the actual CSV file (gzip compressed). As a result, something like this will get printed:
Analysing report from path ./daily-without-ids-00001.csv.gz
Processed 723 lines about EC2 usage.
Time range covered: 2022-08-01 00:00:00 +0000 UTC - 2022-08-22 00:00:00 +0000 UTC (504h0m0s).
REGION INSTANCE TYPE DURATION EMISSIONS
eu-central-1 m4.xlarge 648h0m0s 7.0 kgCO2e
eu-central-1 m5.xlarge 4992h0m0s 66.6 kgCO2e
eu-central-1 t3.large 504h0m0s 3.4 kgCO2e
eu-central-1 t3.micro 504h0m0s 2.5 kgCO2e
eu-central-1 t3.small 72h0m0s 376 gCO2e
eu-west-1 m5.xlarge 4992h0m0s 62.9 kgCO2e
eu-west-1 t2.medium 504h0m0s 3.0 kgCO2e
eu-west-1 t2.micro 1008h0m0s 2.8 kgCO2e
eu-west-1 t3.small 2136h0m0s 10.6 kgCO2e
eu-west-2 m5.xlarge 1512h0m0s 14.5 kgCO2e
eu-west-2 t3.small 480h0m0s 1.8 kgCO2e
TOTAL 175.4 KGCO2EThe output table gives you an aggregation of all EC2 instance usage per region and instance type.
In the last column you get the estimated emissions, expressed as an amount (in g for grams, kg for kilograms, or MT for metric tons) of CO2 equivalents.
The last row contains the sum total of emissions.
In our example above, we see that the input report covers usage from 1st to 18th of August 2022. We see that instances of several types have been run in three different regions.
In order to be able to interpret the result, please read the blog post linked below under Acknowledhememnts. Here is a summary of things to consider.
The power consumption of an EC2 instance has basically been narrowed down experimentally and averaged. The actual power depends heavily on load. We assume that the instance has an average CPU load of 50 percent.
The energy mix and the carbon intensity of the electricity for each AWS region is calculated based on recent yearly averages.
The footprint of machine production is accounted for, based on some reference data and average hardware lifetimes.
Networking and it's electricity usage is not accounted for.
This tool is based on a methodology and data provided by Teads Engineering.
Data in the pkg/footprint folder has been published by Teads under the Creative Commons Attribution 4.0 International License.
Teads provides an interactive web UI for creating estimates along the same lines.
Detail information regarding the methodology can be found in a blog post.