Nick Souter
@nicksouter.bsky.social
Postdoctoral Research Fellow at University of Sussex, studying ways of measuring the carbon footprint of fMRI data preprocessing and analysis 🌱🧠
Big thanks to @loiclnlg.bsky.social, Gabby Samuel, Chris Racey, @lincoln81.bsky.social, Nikhil Bhagwat, Raghav Selvan, and Charlotte Rae for all their work on this paper!
February 8, 2024 at 10:59 AM
Big thanks to @loiclnlg.bsky.social, Gabby Samuel, Chris Racey, @lincoln81.bsky.social, Nikhil Bhagwat, Raghav Selvan, and Charlotte Rae for all their work on this paper!
10. Talk about greener computing
Discussing the carbon footprint of data processing within your neuroimaging community will make it easier to think seriously about this issue. This will help in embedding sustainable ideas into standard practice 🌱
Discussing the carbon footprint of data processing within your neuroimaging community will make it easier to think seriously about this issue. This will help in embedding sustainable ideas into standard practice 🌱
February 8, 2024 at 10:57 AM
10. Talk about greener computing
Discussing the carbon footprint of data processing within your neuroimaging community will make it easier to think seriously about this issue. This will help in embedding sustainable ideas into standard practice 🌱
Discussing the carbon footprint of data processing within your neuroimaging community will make it easier to think seriously about this issue. This will help in embedding sustainable ideas into standard practice 🌱
9. Use existing data
Running analysis on previously collected and preprocessed data is a great way to test hypotheses without using additional energy to preprocess it yourself. We provide a table describing different open access datasets.
Running analysis on previously collected and preprocessed data is a great way to test hypotheses without using additional energy to preprocess it yourself. We provide a table describing different open access datasets.
February 8, 2024 at 10:56 AM
9. Use existing data
Running analysis on previously collected and preprocessed data is a great way to test hypotheses without using additional energy to preprocess it yourself. We provide a table describing different open access datasets.
Running analysis on previously collected and preprocessed data is a great way to test hypotheses without using additional energy to preprocess it yourself. We provide a table describing different open access datasets.
8. Reflect on what needs to be shared
It's good practice to share all raw neuroimaging data on platforms like OpenNeuro. But this storage still has a carbon footprint. Consider whether just sharing preprocessed files would be enough to allow others to make use of your data.
It's good practice to share all raw neuroimaging data on platforms like OpenNeuro. But this storage still has a carbon footprint. Consider whether just sharing preprocessed files would be enough to allow others to make use of your data.
February 8, 2024 at 10:56 AM
8. Reflect on what needs to be shared
It's good practice to share all raw neuroimaging data on platforms like OpenNeuro. But this storage still has a carbon footprint. Consider whether just sharing preprocessed files would be enough to allow others to make use of your data.
It's good practice to share all raw neuroimaging data on platforms like OpenNeuro. But this storage still has a carbon footprint. Consider whether just sharing preprocessed files would be enough to allow others to make use of your data.
7. Push for publicly owned centralised data storage
Large centralised data centres tend to be more efficient, but this often means relying on big tech comapnies to host data. Alternatives like EuropHPC JU may allow storage that is efficient & sustainable:
eurohpc-ju.europa.eu/index_en
Large centralised data centres tend to be more efficient, but this often means relying on big tech comapnies to host data. Alternatives like EuropHPC JU may allow storage that is efficient & sustainable:
eurohpc-ju.europa.eu/index_en
Homepage
The European High Performance Computing Joint Undertaking (EuroHPC JU) is a joint initiative between the EU, European countries and private partners to develop a World Class Supercomputing Ecosystem in Europe.
eurohpc-ju.europa.eu
February 8, 2024 at 10:55 AM
7. Push for publicly owned centralised data storage
Large centralised data centres tend to be more efficient, but this often means relying on big tech comapnies to host data. Alternatives like EuropHPC JU may allow storage that is efficient & sustainable:
eurohpc-ju.europa.eu/index_en
Large centralised data centres tend to be more efficient, but this often means relying on big tech comapnies to host data. Alternatives like EuropHPC JU may allow storage that is efficient & sustainable:
eurohpc-ju.europa.eu/index_en
6. Plan your long-term storage
For files you do need to keep, have a data management plan in place. This could include deleting data after a certain time period (e.g., 10 years) or transitioning it to tape storage.
See spectrum.ieee.org/tape-storage...
For files you do need to keep, have a data management plan in place. This could include deleting data after a certain time period (e.g., 10 years) or transitioning it to tape storage.
See spectrum.ieee.org/tape-storage...
Tape Storage Might Be Computing’s Climate Savior
Tape media generates just 3 percent of the carbon dioxide that hard disks do
spectrum.ieee.org
February 8, 2024 at 10:55 AM
6. Plan your long-term storage
For files you do need to keep, have a data management plan in place. This could include deleting data after a certain time period (e.g., 10 years) or transitioning it to tape storage.
See spectrum.ieee.org/tape-storage...
For files you do need to keep, have a data management plan in place. This could include deleting data after a certain time period (e.g., 10 years) or transitioning it to tape storage.
See spectrum.ieee.org/tape-storage...
5. Tidy up "junk" files
Energy is needed to store and backup data. you can reduce your footprint by deleting files you don't need. For #fMRIPrep, junk files make up to 96% of total size.
Our tool fMRIPrepCleanup, automatically finds/removes this data: github.com/NickESouter/...
Energy is needed to store and backup data. you can reduce your footprint by deleting files you don't need. For #fMRIPrep, junk files make up to 96% of total size.
Our tool fMRIPrepCleanup, automatically finds/removes this data: github.com/NickESouter/...
February 8, 2024 at 10:54 AM
5. Tidy up "junk" files
Energy is needed to store and backup data. you can reduce your footprint by deleting files you don't need. For #fMRIPrep, junk files make up to 96% of total size.
Our tool fMRIPrepCleanup, automatically finds/removes this data: github.com/NickESouter/...
Energy is needed to store and backup data. you can reduce your footprint by deleting files you don't need. For #fMRIPrep, junk files make up to 96% of total size.
Our tool fMRIPrepCleanup, automatically finds/removes this data: github.com/NickESouter/...
4b. Time and location matter
You can also consider choosing cloud computing services based in countries/regions with lower average carbon intensity.
2022 data for countries and US states ⬇️
You can also consider choosing cloud computing services based in countries/regions with lower average carbon intensity.
2022 data for countries and US states ⬇️
February 8, 2024 at 10:53 AM
4b. Time and location matter
You can also consider choosing cloud computing services based in countries/regions with lower average carbon intensity.
2022 data for countries and US states ⬇️
You can also consider choosing cloud computing services based in countries/regions with lower average carbon intensity.
2022 data for countries and US states ⬇️
4a. Time and location matter
The carbon intensity of electricity peaks at busy times and drops overnight and at the weekend. By scheduling preprocessing/analysis to run at low intensity times, you can use the same amount of energy while emitting less carbon.
Data for the UK ⬇️
The carbon intensity of electricity peaks at busy times and drops overnight and at the weekend. By scheduling preprocessing/analysis to run at low intensity times, you can use the same amount of energy while emitting less carbon.
Data for the UK ⬇️
February 8, 2024 at 10:53 AM
4a. Time and location matter
The carbon intensity of electricity peaks at busy times and drops overnight and at the weekend. By scheduling preprocessing/analysis to run at low intensity times, you can use the same amount of energy while emitting less carbon.
Data for the UK ⬇️
The carbon intensity of electricity peaks at busy times and drops overnight and at the weekend. By scheduling preprocessing/analysis to run at low intensity times, you can use the same amount of energy while emitting less carbon.
Data for the UK ⬇️
3. Preprocess conservatively
Reduce compute required for your research by only performing preprocessing steps that are necessary. In our recent study, we provide tips on how to minimise emissions from #fMRIPrep while still getting good quality data:
osf.io/preprints/os...
Reduce compute required for your research by only performing preprocessing steps that are necessary. In our recent study, we provide tips on how to minimise emissions from #fMRIPrep while still getting good quality data:
osf.io/preprints/os...
osf.io
February 8, 2024 at 10:52 AM
3. Preprocess conservatively
Reduce compute required for your research by only performing preprocessing steps that are necessary. In our recent study, we provide tips on how to minimise emissions from #fMRIPrep while still getting good quality data:
osf.io/preprints/os...
Reduce compute required for your research by only performing preprocessing steps that are necessary. In our recent study, we provide tips on how to minimise emissions from #fMRIPrep while still getting good quality data:
osf.io/preprints/os...
2. Track your emissions
There are tools available to estimate emissions from computing, like the Green Algorithms online calculator: calculator.green-algorithms.org
This can be a good first step in understanding the impact of your computing footprint, and taking action to reduce it!
There are tools available to estimate emissions from computing, like the Green Algorithms online calculator: calculator.green-algorithms.org
This can be a good first step in understanding the impact of your computing footprint, and taking action to reduce it!
calculator.green-algorithms.org
February 8, 2024 at 10:51 AM
2. Track your emissions
There are tools available to estimate emissions from computing, like the Green Algorithms online calculator: calculator.green-algorithms.org
This can be a good first step in understanding the impact of your computing footprint, and taking action to reduce it!
There are tools available to estimate emissions from computing, like the Green Algorithms online calculator: calculator.green-algorithms.org
This can be a good first step in understanding the impact of your computing footprint, and taking action to reduce it!
1. Plan and preregister analysis
Repeating data processing or analysis requires additional energy usage, and therefore carbon emissions. Starting with a plan of exactly what you will do with your data can reduce unforseen obstacles, and therefore unecessary repetitions
Repeating data processing or analysis requires additional energy usage, and therefore carbon emissions. Starting with a plan of exactly what you will do with your data can reduce unforseen obstacles, and therefore unecessary repetitions
February 8, 2024 at 10:51 AM
1. Plan and preregister analysis
Repeating data processing or analysis requires additional energy usage, and therefore carbon emissions. Starting with a plan of exactly what you will do with your data can reduce unforseen obstacles, and therefore unecessary repetitions
Repeating data processing or analysis requires additional energy usage, and therefore carbon emissions. Starting with a plan of exactly what you will do with your data can reduce unforseen obstacles, and therefore unecessary repetitions