Computational workloads can now migrate between centralized cloud servers and localized edge nodes in 18.983 seconds while maintaining active mobile connectivity, according to a study published in Frontiers in Computer Science.
A research team from Ikerlan Technology Research Centre and the University of the Basque Country developed FABRIC to solve a persistent infrastructure bottleneck. Current telecommunication deployments keep computational execution separate from core network routing controls. Operators must rely on proprietary tools or static configurations that break when mobile execution contexts shift.
Modern container platforms like Kubernetes orchestrate software across server clusters. However, they lack native mechanisms to tell telecommunication cores to adjust data-plane tunnels toward newly assigned processing endpoints. When latency-sensitive tasks follow mobile users toward regional edge servers, fifth-generation networks demand manual steps to redirect traffic flows.
To bridge this divide, the platform couples workload re-instantiation with dynamic user plane function management. The architecture spans a three-tier execution hierarchy of cloud, edge, and user equipment layers. The system is coordinated by an authoritative Network Resource Model, a Public Land Mobile Network Operator entity, an Edge Computing Service Provider, and an Application Service Provider working as its four primary operational modules.
Standards-Aligned Orchestration Mechanics
Core network management interfaces adhere to 3GPP Technical Specifications 23.501 and 23.502. The framework utilizes a service-based architecture that integrates with containerized network functions. Edge management definitions from TS 28.538 and TS 28.541, along with Provisioning Management Services outlined in TS 28.532, support the service components.
The platform uses standard reference points instead of proprietary messaging formats. The N3 interface connects radio access base stations to user plane functions. The N9 interface bridges multiple user plane functions via GPRS Tunneling Protocol encapsulation.
The system also adheres to ETSI Network Function Virtualization Management and Orchestration standards. This design decouples decision-making logic from physical hardware. It ensures interoperability with existing commercial infrastructure and prospective sixth-generation network paradigms.
Experimental Testbed Validation
The researchers assembled an end-to-end telecommunications testbed running on dedicated hardware to validate the platform. The radio access network used an Amarisoft 5G New Radio stand-alone base station compliant with 3GPP Release 17 specifications. It broadcast across the n78 frequency band.
Core network capabilities ran on an extended instance of the open-source Free5GC software package inside standard containers. Two kernel-based virtual machines served as the computing continuum. The edge server ran on dual Intel Xeon E5-2630 v4 processors with 16 virtual processing cores and 32 gigabytes of memory.
The cloud server utilized an AMD EPYC 9554P processor with 32 virtual processing cores and 64 gigabytes of memory. Both servers operated under Ubuntu 24.04.4 LTS. High-speed wired networking linked the two physical machines together.
Benchmark Migration Latencies
Initial testing used the stress-ng benchmarking utility to generate synthetic compute-intensive workloads. The benchmark ran four CPU stress workers at full target utilization. It included one memory worker maintaining 4,096 megabytes of mapped space and two input-output processes over a 600-second window.
The workload consumed eight equivalent virtual cores on both platforms, resulting in a 32-core cloud server load of 25 percent and a 16-core edge server load of 50 percent.
The initial migration took 18.983 seconds. Step five of the workflow consumed 10.310 seconds to configure and establish the user plane function redirection pathway across the core network.
Across five successive runs, subsequent workload migrations produced a mean total duration of 8.768 seconds accompanied by a standard deviation of 2.117 seconds. Continuous processing overlap kept the initial compute task active until the destination container became operational.
Live Traffic Steering Tests
A second evaluation examined active traffic steering using an external Crosscall Z5 smartphone operating as user equipment. A custom generator transmitted a random integer between 1 and 100 every second via REST API requests directed to a Python collector on the cloud server. The platform then re-instantiated the collector application onto the edge server.
Packet inspection across the N3, N6, and N9 interfaces using Wireshark confirmed continuous data steering. Data packets originating from the user arrived with a fixed source address of 10.60.2.4 while heading toward 10.44.0.1 as their destination. Maintaining continuous packet flow without session disruption, the outer GPRS Tunneling Protocol transport header changed its destination routing from 192.168.9.4 to 192.168.19.157 to re-anchor the active connection.
The time interval separating the final request processed by the cloud instance from the first request handled by the edge node measured exactly 0.578 seconds. Because this transition window remained below the one-second transmission interval of the mobile client, all transmitted values arrived intact. This achieved an absolute 100 percent request success rate.
Total execution time for this single re-instantiation reached 23.746 seconds. The process broke down into 13.123 seconds dedicated to preliminary initialization, 0.008 seconds for step four, 8.492 seconds allocated for user plane path deployment, and 2.123 seconds to finalize operations. Application response times remained between 70 and 98 milliseconds.
Questions About Fabric Operations
How does FABRIC prevent dropped processing cycles during workload migration?
The architecture maintains a continuous processing overlap between instances. The initial compute task remains active until the destination container is fully operational.
What hardware powered the experimental cloud and edge servers?
The edge server ran on dual Intel Xeon E5-2630 v4 processors with 16 virtual cores and 32 gigabytes of memory. The cloud server used an AMD EPYC 9554P processor with 32 virtual cores and 64 gigabytes of memory.
What specifications did the radio access network follow?
The testbed constructed its radio access network using an Amarisoft 5G New Radio stand-alone base station compliant with 3GPP Release 17 specifications.
What was the mean duration for subsequent workload migrations?
Across five successive runs, later workload transfers achieved an average duration of 8.768 seconds with a standard deviation of 2.117 seconds.
Nieto, G., Puerta-Beldarrain, M., Blanco, B., Aldecoa, E., Costa, I., Lopez-Novoa, U., & Liberal, F. (2026). FABRIC: a standards-aligned orchestration platform for compute continuum operations in 5G networks and beyond. Frontiers in Computer Science, 8, Article 1942828.
Related reading