The EXIGENCE consortium recently gathered for its fourth face-to-face meeting in The Hague, Netherlands, hosted by TNO from May 20-22. This three-day gathering proved incredibly valuable, allowing us to sharpen our focus and strategise the next pivotal steps for the coming months. We emerged from a series of insightful workshops and dynamic discussions more aligned and energised than ever.
Key Takeaways from Our Productive Sessions
The initial day was dedicated to a Requirements Consolidation Workshop, a critical step in defining the future energy ecosystem. Our discussions centered on refining scenarios and finalising the functional architecture. Over the project’s first year, we developed fourteen distinct use cases, each with specific requirements, documented in D1.2, “Use Cases and Requirements (Intermediate).” This crucial work will feed into the upcoming D1.4 “Final Deliverable on Use Cases and Requirements” and D1.5 “Final System Architecture,” enabling continued technical progress across all work packages.
Before the meeting, we began generalising and merging these requirements, categorising them as “inter-domain,” “intra-domain,” and “application-level.” Despite these efforts, discrepancies in their formulation, particularly varying abstraction levels, persisted. The workshop directly addressed these issues through an interactive session, with a particular focus on inter-domain requirements. These are crucial as they outline how diverse domains—such as over-the-top video providers, cloud providers, and fixed/mobile network providers—must interact to enable the project’s core objective: end-to-end, service-oriented measurements and optimisations of energy consumption and carbon emissions.
Once again, spending quality time together in person proved exceptionally effective. By the workshop’s conclusion, we successfully defined two top-level, inter-domain requirements, each further broken down into sub-requirements:
- EXIGENCE-enabled domains must expose eco-data for provided services to authorised domains. (This enables historical measurements of energy use and carbon emissions).
- EXIGENCE-enabled domains should expose eco-data expectations for anticipated service requests to authorised domains. (This allows for predictive analysis and cross-domain optimisations).
Day two involved a deep dive into energy metering, covering metrics, measurement methods, trustworthy data collection, and end-to-end service consumption observability. Significant progress was reported across Work Package 3 (WP3) – Energy usage reduction & coordination:
- EVIDEN and Instituto de Telecomunicações (ITAV) presented an initial approach to integrating MANO with K8s (VIMs), focusing on gathering metrics at the cluster/node levels and analysing energy consumption.
- Huawei (HWDU) shared their new publication, which explores how adapting service quality based on temporal carbon intensity variability can lead to additional carbon savings beyond just resource and energy efficiency.
- The National and Kapodistrian University of Athens (NKUA) showcased advancements on Use Case 10, demonstrating significant energy and carbon savings for training AI models on deep edge devices (some with access to green energy) using Federated Learning (FL).
- Athens University of Economics and Business (AUEB) presented their work on user incentivisation for Use Cases 3, 4, and 10. Further elaborated below.
Overall, WP3 has made considerable strides, with several main objectives already achieved. The plan is to continue this excellent work and focus on further publications.
AI and Video Streaming Greener
A particularly engaging session explored incentives for sustainable AI inference, AI training, and video streaming services.
AUEB presented their research on AI inference services, investigating the trade-offs between inference accuracy and response latency. They developed a novel user model that incorporates preferences for accuracy, latency, and environmental consciousness, designing user incentives to reduce associated carbon emissions. This framework includes a Quality of Experience (QoE) model to estimate carbon savings from accepting lower inference accuracy or higher response times, adaptable across various AI application domains and integrating data from other EXIGENCE activities. Their analysis also considered the impact of actual carbon intensity in different countries over daily and monthly scales, assessing the benefits of dynamically routing inference requests to data centers based on location and predicted carbon intensity. These findings were evaluated against real-world service tiers offered by AI inference providers, paving the way for differentiated services that balance performance and sustainability.
In parallel, we are actively investigating incentives for AI training, focusing on optimizing models to lower energy consumption and carbon footprints by leveraging the spatio-temporal dynamics of AI providers’ servers. This approach employs a non-cooperative repeated game to balance rewards against QoE and accuracy in AI training services.
Ongoing work also explores incentives for video streaming services, examining the energy and carbon consumption contributions of each component in the end-to-end delivery chain—including data centers, core and access networks, and end-user devices. This study integrates a QoE model that considers both video quality and the user’s “greenness factor,” reflecting their preference for eco-friendly digital practices like choosing lower video resolutions.
Looking Ahead
The 4th face-to-face meeting in The Hague powerfully reaffirmed the EXIGENCE consortium’s commitment to mitigating the environmental impact of digital technologies. The clarity gained from requirements consolidation, the innovative strides in energy metering, and the work on sustainable AI and video streaming incentives highlights the tangible progress being made.
As we enter the final phase, the work initiated and reinforced in The Hague will be instrumental. The next 12 months are critical for translating these advancements into concrete deliverables that can truly make a difference. We are confident our collaborative efforts will not only contribute significantly to the scientific understanding of digital energy consumption but also lay practical groundwork for a more sustainable, eco-conscious digital future. The EXIGENCE team is energised and fully committed to bringing this impactful project to a successful and influential close.
Stay tuned for more updates as we continue our mission to make digital services greener, one innovation at a time. Thank you once again TNO for your warm welcome and great organisation in making this meeting a success.
Author
Instituto de Telecomunicações
Haoran Chi is a researcher in the Instituto de Telecomunicações, Portugal. He has obtained expertise on 5G (beyond) telecommunication, network management and optimisation, and machine learning. He has plenty of experience in European project management. He is also proactive in scientific community, serving as the Associate Editor of multiple IEEE journals, the Secretory of IEEE IES TC BACM, and Chair/co-Chair of multiple IEEE Standards.
