Greenhouse gas (GHG) emissions have been recognized as accelerators of the Global climate change phenomenon and several human activities take part in it. In particular, the contribution of the computing sector is significant and deemed to grow[1]. While on one side unprecedented results have been reached thanks to the increasing computational power available, on the other, scientific research might become, in the near future, energetically unsustainable. This might be seen as a consequence of the general neglection of the burst of eager requirements/algorithms to carry it on.
In order to avoid underestimating the footprint of Computing, users (as well as data centers) should be able to keep track and analyze the burden associated to their operations, thus making said footprint a viable performance indicator. Equivalently, Reporting such data should be also promoted, in the interest of transparent sharing of computing results and impacts.[2]
We introduce KIG, an open tool written in C++, that allows users and data centers to easily keep track, analyze and report the energy requirements and carbon footprint in gCO2e[3] of their computing tasks. Such tool should help shedding some light on the often not-so-trivial trade- off between performance and environmental Footprint. By gathering detailed data, such tool should also trigger meta-analyses on the behaviour of algorithms as well as computing infrastructures with a view to better leveraging said resources.
To build the library, your machine should have:
- g++ supporting c++17 standard
- cmake (min.required version 3.20)
- pkg-config
- git
package installation varies across different distros. As an example, if you use Aptitude:
sudo apt-get install -y build-essential
sudo apt-get install -y cmake
sudo apt-get install -y pkg-config
sudo apt-get install -y git
The library relies on only one external library, which is provided here:
https://github.com/ToruNiina/toml11 (MIT LICENSED)
we suggest cloning the repo and build the software with cmake.
git clone https://github.com/ToruNiina/toml11
cd toml11
mkdir build
cd build
cmake ..
make all
make install
In order to download KIG, please enter the following commands on your terminal.
mkdir <dir_name>
cd <dir_name>
git init
git clone "<LINK>"
Given the previous step was successfully completed, before installing the library it is advisable to test the build.
cd <KIG>
mkdir build
cd build
cmake ..
make all
It should bring 0 warnings and 0 errors.
Once the build has been tested, installing KIG only requires these two commands:
sudo make install
sudo ldconfig
If you don't want (or you're not allowed) to install the library, there is a containerized solution publicly available on DockerHub.
In order to run KIG from the containerized solution, these are the commands to run it "as an executable":
docker run -dit --name <container_name> --pid host --mount type=bind,source="$(pwd)"/<folder_with_toml_file>,target=/conf <IMAGE_NAME> bash
docker exec -it <container_name> ./KIG_ex $(pgrep -f "<monitored_activity>" | awk 'ORS=" "')
The LaTeX report, if required, will be created in the home of kig_user.
You can also use the container interactively:
docker run -it --name <container_name> --pid host --mount type=bind,source="$(pwd)"/<folder_with_toml_file>,target=/conf <IMAGE_NAME> bash
your user is "kig_user" and your home is "/home/kig_user". There you will find MWE.cpp (and its compiled executable ./KIG_ex). To monitor a process, launch:
./KIG_ex $(pgrep -f "<monitored_activity>" | awk 'ORS=" "')
The LaTeX report, if required, will be created in the home of kig_user.
As a simple guideline for correctly redacting the configuration .toml file, you can refer to the following dummy file:
[owner]
name = "XXX"
title = "YYY"
[infrastructure]
root_folder = "/proc/" #location/naming of /proc folder on your system
cpu_stat_file = "/stat" #location/naming of /stat file in /proc
mem_stat_file = "/status" #location/naming of /status file in /proc
cpu_family = "Skylake" #cpu_family (useful for comparisons)
cpu_tdp = 8 #value in W per-core
n_cpu = 2 #number of usable physical chips on board.
clock_ticks = 100 #output of getconf CLK_TCK, gives equivalence to seconds of clocktick
ram_family = "DDR4" #ram family (useful for comparisons)
ram_freq = 2133 #ideally frequency tells you the wattage required
ram_slots = 1 #active ram slots (optional info)
[energy]
carbon_intensity = 100.0 #Carbon Intensity in your country in g/kWh
power_usage_efficiency = 1.01 #PUE of the cluster/machine running code. Use your national avg if you cannot get detailed data
[1] Jones, N. "How to stop data centres from gobbling up the world's electricity", Nature, 2018.
[2] Lannelongue, L. et al. "Green Algorithms: Quantifying the Carbon Footprint of Computation", Advanced Science, 2021.
[3] Wright, L.A. et al. "Carbon footprinting: towards a universally accepted definition, Carbon Management, 2014.