Track A – 6G-INCOGNICO
Testbed
The well established ATH/ISI Testbed, located in Patras, Greece, is managed by the Industrial Systems Institute (ISI) of the Athena Research Center. It is part of key Stream-C projects coordinated by ISI/ATH (6G-BRICKS and the flagship SUNRISE-6G) as well as Stream-B ones (ADROIT-6G, 6G-INTENSE).
Figure 29. 6G-INCOGNICO Testbed: An overview of the ISI/ATH testbed facility components
Infrastructure
The ISI/ATH testbed, in addition to an end-to-end Private infrastructure, offers a unique cognitive compute continuum 6G enabler, which combines standards compliance, following the new Open Gateway Initiative standards, and cognitive (i.e., AI-driven) operations that aim to fully automate service deployment and resource management. 6G-PATH experimenters can benefit from the deployment of their applications at the ISI/ATH infrastructure, where they can benchmark their execution and measure the KPI improvement from the innovative Cognitive Compute Continuum 6G enabler.
The ISI Testbed (Figure 29) implements a private Cloud-Edge continuum architecture, supporting advanced 5G and B5G experimentation. The hardware includes:
Core Components:
- 3 x 32-Core Servers for 5G Core Network Functions and application workloads
- Multiple Edge and Deep Edge Nodes (e.g., mini-PCs, Raspberry Pi 4)
User Equipment and Access Points:
- OnePlus 8T and Samsung Galaxy S23 Plus smartphones
- Teltonika RUTX50 CPE
- 2 Amarisoft Callbox gNBs (Mini and Advanced)
Radio Capabilities:
- Amari Callbox Mini: 5G SA, 200 Mbps DL / 75 Mbps UL
- Amari Callbox Advanced: 5G SA/NSA, 3 Gbps DL / 2 Gbps UL
The infrastructure is provisioned and managed using Infrastructure as Code (IaC) methodologies and DevSecOps/GitOps principles. Tools used include Ansible, Terraform, Vagrant/libvirt, ServerSpec, and Bash scripting. Each cloud or edge domain is modelled as a Kubernetes (K3s) cluster, with Containers or VMs provisioned as nodes via K3d, Vagrant or KubeVirt. This enables flexible and scalable deployment of containerised applications.
Services
The ISI virtualized infrastructure follows the Cognitive Continuum paradigm, enabling continuous, adaptive decision-making across cloud and edge domains.
Virtualisation, Management and Orchestration
The ISI/ATH testbed offers a Cognitive Compute Continuum PaaS implementation, developed through the ADROIT-6G and 6G-BRICKS projects, which can dynamically commission cloud and edge nodes for containerised application deployment. It is deeply integrated into the private infrastructure, as shown in Figure 30. The ISI/ATH Cognitive Continuum leverages a declarative intent mechanism for requirements specification—an approach also recognised by HEXA-X-II as a critical enabler for future 6G architectures.
This capability is built on Infrastructure as Code, GitOps, and DevSecOps principles. It incorporates Local Decision Agents that ingest telemetry from the Kubernetes control plane and network infrastructure to autonomously adjust resource allocation policies. Decision-making is supported by AI/ML algorithms, including Deep Reinforcement Learning and Workload Prediction, ensuring SLA compliance with minimal resource consumption.
Figure 30. 6G-INCOCNICO Testbed: Overview of the solution design
The Cognitive Continuum is natively integrated with the SUNRISE-6G federation solution, enabling automated application onboarding at both cloud and edge tiers through CAMARA Edge Application Management APIs.
The ISI/ATH testbed also supports multi-region and MEC-style deployments for 6G-PATH experimenters, integrating CAMARA Edge Management APIs with the Cognitive Continuum for end-to-end service assurance.
Specifically, the orchestration system enables:
- Dynamic workload distribution
- Real-time service migration
- Intent-driven optimisation based on live telemetry
To this end, the following tools are used:
- K3d, Vagrant, or KubeVirt for provisioning containers or VMs as Kubernetes nodes
- Lightweight K3s clusters per infrastructure domain
- AI/ML-driven resource orchestration across cloud and edge layers
In addition, the ISI/ATH testbed integrates the Federator functionality of the SUNRISE-6G Operator Platform, enabling automated cross-domain orchestration and federated service deployment. The Federator supports:
- Seamless onboarding and distributing applications across partner testbeds.
- Sharing artefacts (e.g., Helm charts, container images) and resource metadata.
- Multi-site workload migration, coordinated scaling, and SLA enforcement.
- Synchronising CAMARA-compliant APIs and application-level intents across SOP instances.
This federated orchestration layer complements the local decision agents of the ISI/ATH Cognitive Continuum by extending their reach into remote infrastructure domains, enhancing scalability, automation, and intelligence in distributed experimentation environments.
Monitoring
The ISI/ATH testbed implements a comprehensive observability stack that consolidates Prometheus, Grafana, Jaeger, and Kiali within the Execution Lifecycle Management (ELCM) layer. This unified framework enables real-time visibility and control across the entire infrastructure.
Prometheus collects native Kubernetes metrics, such as CPU usage, memory consumption, and pod-level statistics.
Grafana visualises time-series data and supports customisable dashboards and alerting mechanisms.
Jaeger captures distributed traces, providing deep insight into microservice interactions.
Kiali visualises the service mesh topology, enhancing the understanding of service-to-service communication and performance.
Together, these components deliver a single pane of glass for performance monitoring, anomaly detection, and SLA tracking across the platform.
All telemetry—metrics, traces, and service mesh data—is consolidated into a structured SQL-based repository (e.g., PostgreSQL), supporting ad hoc querying, KPI aggregation, and offline machine learning model training. Resulting datasets are curated, anonymised where required, and made openly available to the 6G-PATH community to foster transparency, reproducibility, and further research.
The data collection methodology is structured around the following components:
- Workload Generation:
Synthetic microservice graphs are deployed using the μBench tool to stress compute, storage, and networking layers across the ISI/ATH testbed infrastructure. - Telemetry Collection:
The standardised observability stack captures:
1. Prometheus metrics
2. Jaeger traces
3. Kiali mesh visualisations - Data Storage & Dataset Preparation:
1. Collected telemetry is stored in a structured form (e.g., PostgreSQL) to enable scalable querying, KPI analysis, and model training.
2. Datasets from experimentation campaigns are cleaned, structured, and anonymised, when necessary, then shared with the broader research ecosystem.
This monitoring infrastructure not only supports real-time platform operation but also provides a valuable foundation for AI/ML-driven experimentation, benchmarking, and reproducibility across federated 6G infrastructures.
Security
To ensure trusted and encrypted interconnections, the testbed leverages the pan-European Geant interconnection network through its MD-VPN service, which serves as the backbone of the facility. This backbone guarantees stable, high-throughput connectivity between the federated sites and enables secure onboarding of external testbeds and resource domains.
In addition to network-level security, the ISI/ATH testbed adopts a zero-trust security model for access control and service exposure. All APIs, interfaces, and control-plane services are protected using OAuth2 authorisation protocols and JWT (JSON Web Token)-based authentication. This ensures:
- Fine-grained, identity-based access control for users and services
- Secure token exchange between federated Operator Platforms
- Stateless, scalable access verification aligned with cloud-native security practices
These mechanisms are tightly integrated into the testbed’s federation layer and Operator Platform, aligning with GSMA Operator Platform security principles and enabling secure multi-tenant orchestration within the 6G-PATH ecosystem.
Remote Access
All remote access traffic is routed through the panEuropean GÉANT backbone, providing a resilient, low-latency fabric that interconnects the distributed 6GPATH sites. The following options are offered for remote access:
- SSH-based remote access for project partners.
- Integration of VPN access via OpenVPN or Headscale (self-hosted Tailscale) is under evaluation/deployment to provide secure, identity-based networking across remote clients and edge/cloud nodes.
Key Features and Assets
The 6GPATH testbed blends advanced AI/ML orchestration with Amarisoft-based 5G radios and cloud GPU/FPGA servers, delivering Testing-as-a-Service across the cognitive compute continuum.
1. Cognitive Continuum orchestration
The ISI testbed supports Cognitive Continuum orchestration, enabling AI-driven, seamless integration across devices, edge nodes, and cloud platforms. This approach ensures dynamic workload distribution, low-latency service delivery, and real-time adaptivity by coordinating service components across multiple layers of the compute continuum.
Key characteristics include:
Distributed AI/ML-based orchestration: Leveraging hierarchical decision-making (local + global agents) for optimized service placement, scaling, healing, and migration.
Interoperability across heterogeneous platforms: Supports Kubernetes, O-RAN, and custom MANO stacks, allowing services to move freely between infrastructure domains.
Telemetry-driven intent execution: System behaviour is guided by monitored KPIs and high-level intents (e.g., via CAMARA APIs), enabling predictive and context-aware orchestration.
Cross-domain lifecycle management (LCM): Enables vertical application providers to seamlessly instantiate, monitor, and reconfigure services across sites and vendors.
Support for CAMARA APIs like Traffic Influence, QoD, and Edge Resource Management, enhancing application-level control of network paths, compute locations, and service responsiveness.
This paradigm realises the 6G vision of a unified AI-native infrastructure fabric, supporting Testing-as-a-Service (TaaS) across distributed experimentation facilities with minimal manual configuration and maximum automation.
2. Core & RAN
The ISI/ATH testbed features a versatile and standards-compliant 5G stack that supports both experimental and production-grade use cases.
- 5G Core: The infrastructure integrates both Amarisoft 5GC and Open5GS, enabling flexible deployment scenarios ranging from lightweight standalone cores to feature-rich configurations that support advanced 5G functionalities.
- 5G RAN: The testbed includes two Amarisoft Callbox gNBs—a Callbox Mini for compact standalone setups, and a Callbox Advanced offering full 5G SA/NSA support and higher throughput.
By Q2 2025, the RAN capabilities will be extended with O-RAN-compliant Centralised Units (CUs) and Distributed Units (DUs) provided by Allbesmart and integrated with Ericsson Radio Dot System, enabling conformance with disaggregated RAN architectures. - User Equipment (UEs): The testbed supports a diverse set of 5G SA-compatible user devices, including:
- Two OnePlus 8T 5G smartphones
- Two Samsung Galaxy S23+ smartphones
- One Teltonika TRB500 industrial 5G CPE
- Several Raspberry Pi 4 devices equipped with SIM8200EA-M2 5G HATs, used for lightweight mobile edge workloads and IoT prototyping
This multi-vendor, heterogeneous setup allows researchers to validate both network and application behaviour across real-world and simulated deployment conditions.
3. Private Cloud
The ISI/ATH testbed private cloud consists of three high-performance servers designed to support compute-intensive workloads, virtualization, and GPU-accelerated applications. These servers are interconnected via a high-speed 100 Gbps Nvidia Spectrum-2 SN3420 switch by Melanox, enabling low-latency, high-bandwidth communication across the compute and storage fabric. Shared and local NVMe storage is used to ensure fast data access and scalability.
The first node is a Dell Precision 7865 Tower built around AMD’s 32-core / 64-thread Ryzen Threadripper PRO 5975WX. With 128 GB of ECC memory and a pair of NVMe drives (1 TB for the system and 4 TB for data), it offers more than 4 ½ TB of extremely fast local storage. An RTX 6000 Ada-generation GPU supplies ample CUDA cores for AI training, rendering or heterogeneous scientific workloads. Multiple 10 GbE interfaces—fronted by Intel X710 silicon and aggregated into software bridges—carry container and VM traffic to the 100 GbE spine.
The second node is a Dell Precision 5860 Tower, which balances Intel’s newest core architecture against generous storage. It runs a 24-core / 48-thread Xeon w7-2495X (Sapphire Rapids) paired with 128 GB of ECC RAM. Two 1.9 TB Samsung PM9A1 NVMe drives create a 3.7 TB flash pool for virtual machines and datasets, while an RTX A4500 GPU delivers solid Ampere-class acceleration. Like the first node, it carries a mix of 1-Gb E, Aquantia 10 Gb E and dual-port Intel X710 10 Gb E NICs, giving it plenty of headroom for storage back-plane or overlay-network traffic.
The third physical host is a Dell Precision 7820 Tower built on a 24-core / 48-thread Xeon Gold 6248R (Cascade Lake). This box contains 192 GB of RAM and two 1 TB SK Hynix PC801 NVMe drives that are provisioned as a mirrored pair, leaving roughly 1.9 TB of high-speed storage for production data. Networking is handled by an onboard Intel 1 GbE port and dual 10 GbE Intel X550T controllers, which—together with several 10 GbE software bridges—tie the host into the same fabric as the other two servers. Graphics acceleration is provided by a second RTX A4500, giving the cluster a matched pair of Ampere GPUs for compute or visualisation workflows.
The Mellanox SN3420 switch forms a non-blocking, 100 GbE leaf that interconnects all three servers, ensuring that CPU-to-CPU communication, GPU-direct transfers and NVMe-over-TCP or RoCE storage traffic operate without contention. Spare QSFP28 ports remain available for future nodes or an uplink to a higher-level spine, giving the lab a clear scale-out path while maintaining low latency and high throughput across its compute, GPU and storage resources.
4. Operator Platform
The ISI testbed integrates with the SUNRISE-6G Operator Platform (OP), which adheres to GSMA Operator Platform standards and supports several core roles—Capabilities Exposure, Federation Manager, and Service Resource Manager (SRM), each delivering key functionalities that ensure seamless service delivery across federated Operator Platforms. Interoperability with 3GPP Capability Exposure Interfaces (including NEF and CAPIF) is ensured via the Southbound Interface (SBI), enabling native interaction with 5G Core networks and infrastructure platforms. All OP components are implemented as open-source software and mapped to the functional roles defined by GSMA OPG, facilitating interoperability, standardization, and scalable federation across Europe’s 6G testbeds.
Infrastructure Overview
| Category | Asset | Details / Specs |
|---|---|---|
Compute Servers | Dell Precision 7865 | AMD Threadripper PRO 5975WX, 128 GB RAM, 5 TB NVMe, RTX 6000 Ada |
Dell Precision 5860 | Intel Xeon w7-2495X, 128 GB RAM, 3.7 TB NVMe, RTX A4500 | |
Dell Precision 7820 | Intel Xeon Gold 6248R, 192 GB RAM, 1.9 TB NVMe, RTX A4500 | |
Networking | Mellanox SN3420 Switch | 32× 100 GbE ports, Spectrum-2 ASIC, Cumulus Linux / SONiC, VXLAN/RoCE support |
MikroTik Router | Handles edge routing, NAT, firewall, WAN gateway | |
Power | APC UPS | Battery backup and surge protection for core infrastructure |
Radio Test Equipment | Amarisoft Callbox Mini | 5G SA, 200 Mbps DL / 75 Mbps UL |
Amarisoft Callbox Advanced | 5G SA/NSA, 3 Gbps DL / 2 Gbps UL — full-featured test gNB | |
User Equipment (UE) | OnePlus 8T | 5G NR smartphone — UE for testing, logging, throughput validation |
Samsung Galaxy S23 Plus | 5G smartphone with advanced RF capabilities | |
Teltonika RUTX50 | Industrial 5G router/CPE — supports 5G SA/NSA, GNSS, dual SIM, WiFi 5 | |
Access Points | Amarisoft gNBs (Mini + Adv.) | Act as 5G SA/NSA base stations for RAN testing and UE connectivity |
Edge Nodes | Mini PCs | Compact x86 platforms for edge/server roles in use cases |
Raspberry Pi 4 | ARM-based edge nodes for lightweight workloads, monitoring, and IoT integration | |
XR Devices | Meta Quest 3 | Mixed Reality headset for immersive app testing, spatial UX, and XR development |
Targeted Use Cases
- All verticals are accepted
- All use case topics are accepted. Testbed prefers use cases targeting one or more of the following topics:
- Immersive XR Gaming & Metaverse Experiences.
- AI-Assisted Remote Industrial Inspection.
- Real Time Digital Twin for Smart Ports.
- Public Safety & Disaster Response B5G Cells.
Predictive Edge Analytics for Connected Mobility.