Environmental Sustainability
Energy use has recently become a limiting factor for HPC due to the difficulty of procuring MW, cost and cooling issues. It is, however, just the tip of the iceberg as, in the current context of global warming, collapse of biodiversity and gradually exceeding planetary limits, environmental impacts of IT are significant and need to be reduced at the European scale to fulfil the Paris agreement. Conversely, large science projects, in the HEP and RA domains for instance, call for increased use and growth of HPC infrastructures and will face the challenge of doing more with fewer impacts to be environmentally, politically and socially acceptable. This is not simply a matter of improving the efficiency of the processes, although this will help, but of gaining control over the different impacts of IT in order to allow it to fit in schemes to globally reduce mankind's impacts. A holistic approach is required as science moves into the era of big data and because of the exponential adoption of AI to ensure scientific computing can continue responsibly in spite of the growing environmental pressure.
Strategic Rationale and Evidence Base
Strategic Importance
Energy use has recently become a limiting factor for HPC due to the difficulty of procuring MW, cost and cooling issues. It is, however, just the tip of the iceberg as, in the current context of global warming, collapse of biodiversity and gradually exceeding planetary limits, environmental impacts of IT are significant and need to be reduced at the European scale to fulfil the Paris agreement. Conversely, large science projects, in the HEP and RA domains for instance, call for increased use and growth of HPC infrastructures and will face the challenge of doing more with fewer impacts to be environmentally, politically and socially acceptable. This is not simply a matter of improving the efficiency of the processes, although this will help, but of gaining control over the different impacts of IT in order to allow it to fit in schemes to globally reduce mankind's impacts. A holistic approach is required as science moves into the era of big data and because of the exponential adoption of AI to ensure scientific computing can continue responsibly in spite of the growing environmental pressure.
Current State
The energy challenge and GHG emissions of datacentres have been gaining traction and moves towards the use of renewable energy and more efficient cooling have been taken. While this allows for reducing the environmental impacts of the use phase, it does not allow for addressing them globally. Life-cycle assessment methodologies (ISO 14040, ISO 14044, product environmental footprint) and databases, from the cradle to the grave, have been emerging to quantify the impacts and improve the design, fabrication and overall management of the IT resources, as well as avoid the transfer of impacts between phases, impact types or geographical regions. To date, however, tooling and policies to forecast and manage impacts are still in their infancy and coordination is lacking to gain control on the whole and first stall their growth, then reduce them overall. Impacts can be local (e.g. water use) or global (e.g. climate change) and are generated by diverse communities (end users, datacentre managers, hardware suppliers, waste managers etc.) at the national or international scales, which makes them complex to address consistently. In a similar way, preliminary mitigation of negative social impacts is under development with approaches based on the more recent ISO 14075 norm.
Community Evidence
The optimisation paradigm adopted by RA and HEP is centred on weak scaling: the ability to handle larger problems using a larger amount of resources. In this frame, section 2.5 gap #10 in the Technical Blueprint identifies the need to monitor workflows, performance and energy usage across sites to efficiently exploit a multi-infrastructure compute and storage continuum. This focus on efficiency, as it allows for pushing the limits of what is technically feasible, has and will lead to further increases of storage, network and computing resources in contradiction with the imperative to reduce the environmental impacts. This is something known as the rebound effect or Jevon's paradox.
The D6.1 Technical Blueprint identifies a number of gaps relevant to environmental sustainability. Gap #10 in section 2.5 resulting from WP5 analysis relates to monitoring and observability with the lack of consistent, cross-site monitoring of workflows, performance, and energy usage (D6.1). This forms part of the inefficient resource usage gap in section 3. The same section points out that execution is carried out irrespective of environmental impacts which implies that the latter are only determined a posteriori, ie. too late to be avoided or reduced. A third gap in this section corresponds to significant digital waste due to poor data and software reuse and of failing software. The fragility of operations is finally pointed out, which in the light of the increasing pressure on energy and cooling is susceptible of leading to increased waste or the inability to operate due to shortages of resources.
Technical Foundation
Sections 3.1 and 3.2 of the Technical Blueprint propose a path for reconciling weak-scaling optimisation efforts with environmental sustainability through the development of tools allowing to view them through the prism of environmental impacts. These provide a technical basis for sustainability policies to be enforced at the continuum and datacentre levels to address their impacts with a holistic approach. Central to this approach is using environmental impacts as credits allocated to users at all levels. This will make them visible and promote responsible behaviours (as opposed to abstract quantities like core hours). It will also allow for framing their expenses within a global environmental impact envelope defined by higher-level policy-makers so that the environment is indeed protected. It is expected that datacentres will arbitrate between procurement, maintenance and usage impact expenses on this basis. In doing so, the latter will flow down to end users who will then need to optimise and prioritise their uses accordingly. While the tooling is end-user oriented, these policies are intended to be set by datacentres, governing bodies and higher-level policy-makers in order define a large-scale comprehensive framework for the management and use of European IT infrastructures within determined environmental impact envelopes. Guidelines for defining these policies are provided below but precise definitions based on metrics and setting thresholds is beyond the scope of the project as they are the responsibility of datacentres, governing bodies and high-level policy makers, each at his own level.
In the light of the gaps summarised above, the D6.1 Technical Blueprint recommends the development of software-hardware optimisation to optimally use resources and distribute data and compute, tailoring execution to sustainability in terms of forecasting impacts and providing end users with means to map and prioritise their jobs so they can be scheduled in an impact-aware manner, FAIR data and software publication, in direct relation to Priority P05 (Data Management Continuum), fine software quality assessment prior to the use of production-level resources to reduce the probability and consequences of failures and improvements in the IT’s adaptation capabilities, notably through the resilience of software execution to varying or failing hardware resources.
This approach is complemented with policies addressing the other phases of the IT lifecycle.
European Alignment
The European Green Deal aims to make Europe the first climate-neutral continent. This is a significant ambition, a top-level objective addressing climate change globally across the union. Recommendations for technical developments in the Technical Blueprint and policies in this document aim to empower policy-makers in this sense and allow for this objective to flow down in quantitative terms to datacentres and end-users.
EuroHPC JU Strategic Commitment: Europe has demonstrated leadership in sustainable HPC through:
- JUPITER JEDI #1 Green500 ranking
- MareNostrum 5 PUE < 1.08 industry-leading efficiency
- SEANERGYS €16.8M EU investment in energy-efficient workload management
EOSC and Broader European Strategies: Environmental sustainability aligns with broader European strategies. The Digital Decade Policy Programme 2030 emphasises sustainable digitalisation, and EOSC principles include environmental responsibility for research infrastructures.
JENA Federated Computing Vision: JENA Computing White Paper positions environmental sustainability as one of the central challenges for federated computing, requiring transparency, lifecycle optimisation, and cultural shift towards sustainable computing practices.
Target Outcomes and Impact Assessment
Strategic Outcomes
1. Forecast and Reporting of Environmental Impacts: Operational capability to simulate, forecast and report environmental impacts of computing workloads, enabling informed resource allocation and scheduling decisions.
2. Improved Software Quality: Software development practices that prioritise energy efficiency and resource optimisation alongside performance and scientific accuracy, contributions to libraries to increase the lifetime of software, help it mature and standardise practices to facilitate co-design.
3. Improved Efficiency of Execution: Multi-objective optimisation of computing workloads balancing performance with environmental impact (carbon emissions, energy consumption, resource utilisation) (D6.1).
4. Lifecycle Extension: Procurement and operational policies that extend hardware lifetimes, accounting for fabrication impacts (upstream) and end-of-life disposal (downstream). In the European energy mix, fabrication phase outweighs usage phase, making replacement for efficiency gains rarely beneficial (D6.1).
Success Indicators
1. Environmental Transparency: Environmental monitoring (impact budgets, PUE/CUE etc.) operational at major European datacentres with public reporting
2. User Engagement: Sustainability metrics visible through dashboards to users during job submission
3. Procurement Integration: Sustainability metrics (embedded carbon, water usage, repairability, HEPscore/watt etc.) adopted as standard procurement criteria alongside performance metrics
4. Lifecycle Extension: Hardware operational lifetimes extended across SPECTRUM infrastructures
5. Cultural Shift: Training programmes on sustainable computing practices operational, awareness to environmental impacts and ecologically friendly ways of designing and using IT infrastructures through communities of practice
6. Policy Alignment: Integrated sustainability policies, accounting and management from the projects to the infrastructures with coordinated and overarching supervision by European HEP/RA institutions and EuroHPC.
European Impact
Europe already leads in green HPC, with JUPITER JEDI ranked #1 on the Green500 and MareNostrum 5 achieving PUE below 1.08. Reducing GHG emissions from research infrastructures contributes to European climate neutrality by 2050 and ensures continued public support for large-scale investments such as HL-LHC and SKA. Energy efficiency improvements also reduce operational costs, while lifecycle extension provides greater economic and environmental benefit than frequent hardware replacement (D6.1).
Implementation Pathway and Stakeholder Coordination
The Technical Blueprint recommends a number of tools to support the approach described here. Although part of these tools have been the focus of research activities, achieving the reliability and efficiency required for these tools to be used in production, that is for the effective management of IT resources, requires this effort to be intensified on the short term before the tools can be made mature through consolidation based on their use through an experimental stage described as part of the phases below.
Implementation Phases
Given the European scientific and computational ambitions, the Paris agreement and the corresponding European Union policies call for immediate action with the available means and for their extension. The paragraphs below describe the programmatic aspects while additional detail useful to make this actionable is provided in Annex C.3.
- Short-term (Years 1-3) : This first phase has two facets. One consists in putting in practice available solutions to achieve a reduction of IT’s environmental impacts and provide some relief to the environmental crises in the short term with limited disruption. The other is preparatory to further this approach. Its purpose is to plan, develop and experiment with the solutions required for the change of paradigm fully integrating the consideration of environmental impacts as a constraining aspect for the development, management and use of the data and compute continuum.
- Medium-term (Years 3-5) : In the medium term, as a result of the experimentations carried out in the first phase, the different measures should be consolidated and put into wide-scale practice. It will also be time for experimenting with the more complex measures for which the first phase has laid a basis.
- Long-term (Years 5+) : Given the urgency of the environmental crises and the expected growth of IT, all of the different proposed measures (or adaptations thereof) should be consolidated and put into practice in this time frame. Given the likely continued rapid evolution of IT, this will also require continuously monitoring their effectiveness and adapting them to the hardware and software systems, as well as to the environmental situation.
Stakeholder Roles
- Scientific Communities - Responsibilities: train users on software development for low environmental impacts and mindful use of resources (priorities, experimentation plan) – Benefits: reduction of digital waste, resilient and adaptive execution, improved software quality
- Research Infrastructures - Responsibilities: contribute to sustainability governance, contribute to data and software reuse, incorporate IT environmental impacts in the wider assessment of those of the research infrastructure - Benefits: improved coordination between user communities, detailed accounting of environmental impacts resulting from computing and storage
- Computing & Data Service Providers / e-Infrastructures - Responsibilities: contribute to sustainability governance, manage the overall environmental impacts of infrastructures, allocate end-user shares such impacts as credits, support users and policy makers in tracking and managing them – Benefits: sustainable operations, detailed accounting of environmental impacts
- Policy Makers / Funding Bodies - Responsibilities: define environmental impact budget at the scale of the compute and storage continuum, evaluate sustainability governance - Benefits: meet obligations resulting from the Paris agreement, sustainable operations,s detailed accounting of environmental impacts
Resource Needs
Given the importance of embedded impacts resulting from the fabrication of IT equipment, the transition to a more sustainable management of IT infrastructures throughout their life-cycle relies primarily on the involvement of personnel, both in a dedicated way as part of the sustainability governance staff and the development of the underlying tooling and in terms of specialisation in eco-design of infrastructures and software at the datacentre and end-user levels.
The development of the tooling recommended as part of the Technical Blueprint, although it will build on existing solutions, will require a significant amount of effort to be tailored to environmental impacts. This tooling consists mostly in software.
Critical Dependencies
The management of environmental impacts involves actors at all levels of the IT hierarchy and is by nature transverse as it involves all phases of the life cycle, hardware and software. A holistic, systems engineering view is required to avoid excessive disruption as well as the transfer of impacts. For this reason, coordination is seen as a critical dependency for this effort to be successful.