CELTIC-NEXT AI-NET ANIARA Highlights

Compute and AI enablers for an intelligent and sustainable 6G platform


Ali Balador
Ericsson

Modern society expects increasingly ad­-vanced digital applications to be available anytime and anywhere. To make future mobile networks ready in terms of performance and functionality for use-case scenarios like Smart Cities and Smart Manufacturing, new computation and Artificial Intelligence (AI) enablers need to be provided for future mobile networks to transform them into a communication platform offering services that go beyond communication.

AI-NET-ANIARA is a flagship CELTIC-NEXT project where three EU countries were involved including Sweden, Germany, and UK. The project ended in February 2024. This innovative project provided several technology enablers and showed exciting demonstrations of these technologies that will change the world of communication. The project goal was to complement the evolution of 5G with crucial technical enablers towards an intelligent and sustainable 6G platform, offering services beyond pure communication, including compute & AI.

Project results:

› Carrier grade AI technologies for telecom edge automation, and intelligent management in support of services with guarantees on performance and energy consumption.
› Novel AI concepts such as robust and reliable federated learning, reinforcement learning for service mesh, intelligent feature selection, and transfer learning.
› MLOps infrastructure for AI. ANIARA extended the Feature store solution by ANIARA partner Hopsworks to support real-time model serving all the way to the edge.
› Execution runtime to process edge applications using WebAssembly. A dynamic computational offloading solution was developed and its potential as a future 6G service has successfully been demonstrated on the SmartCity application by ANIARA partner Univrses.
› A modifed Kubernetes scheduler enabling the allocation of workloads based on the energy state of nodes and/or devices. This modification improved the energy efficiency of the system. Furthermore, the project designed a new resource scheduling mechanism for Kubernetes that can closely match the time-varying traffic profile of users. This led to a 30% improvement in resource efficiency.
› Designed, built, and operated two versions of Aniara Edge node data center demonstrators. These demonstrators include a fresh air cooling approach, where the cooling system is highly integrated in the Edge node design. An intelligent power system ensures that the servers are provided with enough power also during periods of limited external power supply and a photovoltaic system to reduce the amount of power from the grid, improve the sustainability and maximise robustness.
› Developed new edge processing devices deployed in Stellantis manufacturing sites that monitor energy consumption and machine condition. This data is then transmitted wirelessly to, for example, optimize the energy consumption in the paint shop.

Conclusions:

CELTIC-NEXT project AI-NET ANIARA provides technical enablers for an intelligent and sustainable 6G platform in line with the ITU-R IMT 2030 recommendations. AI-NET ANIARA project designs and builds robust and energy efficient mini edge datacenter infrastructure. For efficient use of distributed edge resource, improvements presented the existing resource orchestration methods. Moreover, to support automated operations of 6G networks and their functions, several studies have been performed on scalable distributed intelligence methods. Finally, AI-NET ANIARA contributed with innovative service ideas that go beyond pure communication, increasing future network platforms usage vector by compute & AI services exposed to applications. These include privacy preserving AI/data services, a real-time online features store for edge applications, as well as a dynamic device offloading service to improve the experience of mobile devices and apps.

Overall, AI-NET-ANIARA was successfully living up to its project promises of accelerating digital transformation by the efficient use of a highly integrated and flexible edge infrastructure that is programmable across all its components.


Figure 1: Dynamic device offloading for a SmartCity app


Figure 2: ANIARA Edge data center


Figure 3: Closer look of the stack

CELTIC-NEXT website – https://www.celticnext.eu/