The evolution of the edge-to-cloud continuum brings transformative potential, but also significant hurdles in distributed data management. The European project ACES is addressing these challenges through a cognitive-by-design methodology, moving beyond simple data storage toward active knowledge interpretation.
A core innovation in ACES is the introduction of a resource pool, which presents a significant shift from current edge definitions. By leveraging technologies like CXL and PCIe switches, the system can request single resources for pod processing rather than relying solely on fixed node capacities. This innovation prevents resource exhaustion and stability issues, though orchestrating these dynamic bindings remains a primary challenge as traditional scheduling techniques struggle with such complexity.
The management of data is further complicated by the diverse requirements of modern services. Long-standing running applications require iterative in-memory computing, while batch and stream processing demand regular intervals or real-time responses. As future workloads become increasingly interconnected, the system must navigate the intricate relationships among pods. This is where the need for swarm agents becomes critical. These represent related pod splits from a specific service that must be managed collectively. Currently, orchestrators often overlook these relationships, but ACES recognizes that placing interacting microservices or database-dependent pods in closer proximity can significantly reduce latency and enhance performance.
To solve data heterogeneity, ACES utilizes a graph-based data model aligned with NGSI-LD (Next Generation Service Interfaces – Linked Data). By transforming data from diverse sources, like TimescaleDB and Neo4j, into a unified semantic format, the project ensures that machines understand the context of metrics and alerts. This is supported by NATS-based pipelines and the use of temporal knowledge graphs. Unlike static graphs, these graphs track how relationships change over time, which is vital for detecting anomalies in dynamic microservice environments. While challenges like temporal synchronization and GNN scalability persist, ACES’s modular architecture provides a scalable foundation for autonomous edge intelligence and transparent data governance across multi-stakeholder environments.
Project deliverable
Find more information in Deliverable 4.5 “ACES – Distributed knowledge base and data management systems” from the official project website.
Publications
M. Schranz, K Harshina, P. Forgacs, and F. Buining: Agent-based Modeling in the Edge Continuum using Swarm Intelligence. In Proc. International Conference on Adaptive and Self-Adaptive Systems and Applications, April 2024.
M Gojković and M. Schranz: Bottom-Up Resource Orchestration in Edge Computing: A Pod Profile-Aware Agent-Based Approach, CPS Summer School PhD Workshop, Alghero, Sardinia, Italy, September 2025.