Sustainable computing¶

Computational research impacts the environment through the energy used as well as the carbon emissions and other environmental impacts associated with the production and disposal of computing hardware. Many researchers are increasingly interested in how to reduce the environmental impacts of their computational research, and some funders, such as the Wellcome Trust, require grant recipients to state how they will assess and tackle these impacts.

This page provides resources to learn about the environmental impacts of computing, and some approaches to assess and reduce the impact of your research.

Sources of carbon emissions in computing¶

Greenhouse gas emissions are typically split into three categories. Typcially, only Scope 2 and Scope 3 emissions are relevant for research computing.

Scope 1: direct emissions from the service or organisation, such as on-site fuel combustion or fleet vehicles. There is nothing of this category currently associated with the CREATE HPC service at King's.
Scope 2: indirect emissions related to the generation of purchased energy, such as heat or electricity. For CREATE HPC, this would correspond to any emissions from the electricity used by the service. The energy contract for the data centres we use is based on 100% certified renewable energy, so the scope 2 emissions of CREATE HPC can be reported as zero.
Scope 3: Other indirect emissions, such as from an organisation's supply chain. For CREATE HPC, this corresponds to the embodied carbon emissions of its hardware and supporting physical infrastructure, i.e. emissions associated with their manufacture, shipping, and disposal.

Principles for reducing carbon emissions¶

The Green Software Foundation defines three "Green Software Principles" that can reduce the carbon emissions of software:

Energy efficiency: consume the least amount of electricity possible
Hardware efficiency: use the least amount of embodied carbon possible
Carbon awareness: do more when electricity supplies are coming from renewable sources, and less at times when electricity is generated from carbon-intensive sources

Computational researchers often don't have control over the supply chain of the hardware they use or the electricity sources used to power their research computing. However, there are things all researchers can consider to reduce the emissions associated with their research.

Energy efficiency¶

Make your code more efficient: this can be complicated, but for code that will be run many times, it's worth investigating whether you can make it run more efficiently. Many programming languages have tools for profiling your code to identify which operations take the most time.

Use more efficient tools or versions: often, newer tools are more efficient. This can include using newer versions of the same tool or even programming language.

Reduce the number of times code needs to be re-run: it's more efficient to test your code on a small test dataset and ensure it's working correctly, than to run it directly on a large dataset only to discover a bug. Running small-scale tests to identify issues faster also has the benefit of saving you time!

Hardware efficiency¶

Hardware efficiency includes making the most use out of existing hardware, rather than having to buy more or newer hardware.

Make sure your HPC jobs don't request more resources than they need: you can use tools like sacct to check whether your HPC jobs used all the cores/threads and all the memory you requested. Requesting only the resources you need will ensure that HPC resources are allocated effectively (and has the bonus that if your jobs request fewer resources, they'll likely spend less time waiting in the queue!).

Be strategic around what data you keep: consider whether you can delete intermediate files in a data analysis pipeline, or raw data that is backed up elsewhere such as in a public data repository. While you don't want to have to frequently re-run analyses to regenerate results (as this would use more energy!), there may be intermediate files you can safely delete.

Prefer using existing hardware to buying new hardware: to make the best use of existing HPC hardware, it should be in use as much as possible. Therefore, rather than buying a new server or high-powered desktop for your lab or department, consider whether you can use the existing CREATE HPC or CREATE Cloud resources instead.

Extend the lifetime of existing hardware: make use of warranties to get hardware repaired rather than replaced, where possible. Consider whether each new research group member needs their own new laptop/desktop, or whether there's an unused machine that could be refurbished and repurposed - if you're using the HPC for intensive research computing, you probably don't need the latest model of laptop.

Carbon awareness¶

The carbon intensity of electricity production varies over time and across different locations, depending on the sources of energy, with renewable sources such as wind and solar being low carbon, and fossil fuel sources having higher carbon intensity. Research computing is typically carried out on hardware connected to the national grid, which is supplied by a mix of sources, although data centres may have an electricity contract which uses 100% certified renewable energy. As a researcher, it's likely you don't have much choice over the carbon intensity of the electricity used to power your research computing. However, there are a couple of things you can consider:

Choose supercomputing resources based on carbon intensity: if you are using UK or international supercomputing facilities, consider choosing one that is located in a region with low carbon intensity electricity suuplies. For example, ARCHER2 is located in Scotland, where wind farms are a major source of electricity generation.

Carbon-aware job scheduling: if you run your own server, tools like the Climate-Aware Task Scheduler can help schedule jobs at times when carbon intensity of the national grid is lower. This type of scheduling is most useful on a system that is not used at full capacity - on systems with high usage, it's best to have continuous usage to make best use of the hardware and minimise embodied carbon per job.

Green DiSC¶

Green DiSC is a new certification scheme which provides a roadmap for research groups and institutions who want to tackle the environmental impacts of their computing activities. Both individual groups and central teams can apply for Green DiSC certification. The e-Research and the RMID Sustainability in Research teams are working together on a central team application. If you are interested in applying as a research group or to find out more, see the Sustainability Champions Sharepoint page.

Resources¶

Green Software for Practitioners, a training course from the Green Software Foundation and Linux Foundation.

GREENER principles for environmentally sustainable computational science, Lannelongue et al., 2023.

Ten simple rules to make your computing more environmentally sustainable, Lannelongue et al., 2021.

The Green Algorithms calculator is a tool to estimate the carbon footprint of your computations.

Best Practices in Green Software- Principles, Patterns and Tools, Colin Sauze & Sadie Bartholomew, presented at the Software Sustainability Institute Collaborations Workshop 2024.