Open End-to-End Platform Infrastructure


  • The research community needs an open platform to research and experiment end-to-end, from edge services to RAN through the mobile core to services and orchestration.

  • Intel worked with partners at the Open Networking Foundation (ONF) to develop Aether, which is recognized as one of the leading open source 5G platforms.

  • The open platform infrastructure enables researchers to research and develop innovative edge services, AI/ML based end-to-end quality of service and resource allocation, intent-driven orchestration, and security.



Vision and Motivation

Over the last decade, an evolution in architecture such as software defined networks (SDNs) and network function virtualization (NFV) led to an end-to-end redesign of networks, from open radio access networks (RANs) to mobile cores running on high-volume server platforms hosted either on private, public, or hybrid clouds. The cellular network evolved to 5G, enabling distributed edges to run local edge services to preserve service data locality, improve throughput and latency, and optimize energy efficiency at the edges of the network. The advent of artificial intelligence/machine learning (AI/ML) enables the addition of intelligence in the network through innovative edge services, open RAN, and mobile core optimizations.

The research community needs an open platform to research and experiment end-to-end, from edge services to RAN through the mobile core to services and orchestration, and deployment of these functionalities in cloud environments. This transparency allows developers to understand the challenges associated with scalability, security, and end-to-end quality of service (QoS).


Figure 1. End-to-end network from edge services to the data center.

To address this need for an open infrastructure platform, Intel worked with partners at the Open Networking Foundation (ONF) to develop Aether, which is recognized as one of the leading open source 5G platforms. Curated open source mobile infrastructure software provides customers with industry-tested solutions that are smart, simple, and secure. The end-to-end solution can be deployed with a range of RAN options, and its highly scalable nature enables users to start with small deployments and scale as requirements expand into large distributed networks supporting 5G connectivity. The project transferred from ONF to the Linux Foundation in December 2023.

Open Platform Infrastructure Development

The project stemmed from close collaboration between Intel, AT&T, ONF, and the academic community through an Intel Sciences and Technology Center (ISTC) focused on software defined networks and network function virtualization. Partners transformed the traditional central office running on dedicated devices into a Central Office Re-Architected as Datacenter (CORD) running on off-the-shelf servers.

This research led the partners to develop the Open Mobile Evolved Core (OMEC), a re-architected 4G/LTE mobile core using the 3GPP control and user plane separation (CUPS) model that follows SDN/NFV architecture principles. OMEC uses a Data Plane Development Kit (DPDK) based data plane to sustain high throughput and specific features of the Intel Architecture CPU, such as Intel® Software Guard Extensions (Intel® SGX), a secure enclave to protect lawful intercept interfaces as well as charging data records (CDRs). Led by Intel, OMEC’s development was released to the ONF community and ONF announced the first production roll-out of OMEC in Poland.

Real-World Use Cases

From there, the work continued as Intel, ONF, and other partners evolved the mobile core to a cloud native 5G mobile core. In 2021, ONF celebrated the first field trial implementing fully disaggregated open RAN solutions in Berlin, Germany. The live trial leveraged the open mobile platform and featured horizontally disaggregated hardware – separate radio unit (RU), decentralized unit (DU), and centralized unit (CU) – and vertically disaggregated software components, including an open source near real-time RAN information controller (RIC) and xApps from ONF’s SD-RAN project. Overall, the trial presented a foundational model for future open RAN deployments.

Figure 2. Open mobile 5G platform infrastructure.

Intel, Radisys, and Foxconn collaborated closely with du from the Emirates Integrated Telecommunications Company, to enable AI-based edge services running on the open mobile infrastructure 5G platform to be deployed for proof-of-concept in du’s 5G lab in Dubai. Researchers at Intel Labs developed edge services for anomaly detection of defects in solar panels in large solar fields and a visual data management system (VDMS) to detect unattended luggage.

The Aether platform is also the basis for the Defense Advanced Research Projects Agency’s (DARPA) Pronto research project at Stanford University, Princeton University, and Cornell University. The contract provided DARPA's project with ONF's Aether software as part of a research and secure 5G network infrastructure project.

Research on the Open Platform Infrastructure

The open platform infrastructure enables researchers to research, develop, and experiment on innovative edge services, AI/ML based end-to-end QoS and resource allocation, and intent-driven orchestration (IDO) as well as ensuring security.

Innovative edge services: Few edge services exhibiting different requirements from high throughput to low latency have been deployed on the open platform infrastructure and benefited from edge processing properties such as lower latency, higher throughput, and data locality to preserve privacy.

  • AI/ML based defect detection: After learning in an unsupervised manner about object features from anomaly-free training data, this edge service can then identify anomalies in objects. This usage model requires high throughput to transfer video images of objects to the edge service to check for defects as well as low latency when the service warns the user/application of a defect detected.
  • Video data management system: This edge service is designed to enable efficient access to visual data. The VDMS uses a custom library to store and retrieve visual data, providing an interface for machine-friendly and traditional formats to accept complex queries to retrieve images or video segments. This usage model requires high throughput.
  • Autonomous mobile robot (AMR): This usage model partitions the processing traditionally done on board the robot between the robot and the edge network. Enabling robots to offload computing workloads to the edge where higher compute capabilities reside enables advanced intelligence to constrain mobile robots, reduce latency by accelerating workloads on the edge as well as optimize robot battery life and cost. This usage model requires tight latency provided by the support of quality of service to meet ultra reliable and low latency communication (URLLC) traffic requirements.

AI/ML based RAN optimizations: At the RAN level, these various xApp small applications run on an ORAN RIC, and control functionality and operation of the RAN. Many have been developed and, in some cases, deployed with the following operator partners:

  • 5G system energy efficiency xApp, an AI/ML based xApp controlling when to turn on/off cell or multiple-input and multiple-output (MIMO) chains based on traffic workload while preserving end user service level agreement.
  • Graph neural network (GNN) based connection management xApp load balancing user element (UE) connection to the most appropriate cell to optimize user and network operations.
  • To learn more, visit the Intel Network Builders Document Library.

AI/ML based resource allocation: At the mobile core level, Bayesian optimization (BO) is used to learn traffic characteristics and the run time using Intel architecture platform features. The Intel® Speed Select Technology (Intel® SST) allocates the right amount of CPU resources such as the operating frequency. The Intel® Resource Director Technology (Intel® RDT) allocates the right amount of shared resources such as cache, memory, or I/O bandwidth to the concerned process, virtual machine, or container running the workload in order to meet the end user service level agreement. Figure 3 shows how using BO techniques on a 5G UPF we can dynamically adjust the CPU frequency to process the incoming packet rate without losing packets and achieve a 42% power reduction.

Figure 3. Using BO techniques on a 5G UPF, the CPU frequency can be dynamically adjusted to process the incoming packet rate.

Intent-driven orchestration (IDO): End-to-end AI/ML based automation is provided with the use of intent-driven orchestration (IDO) enabling the end user to define high-level intents without needing knowledge about the underlying platform and its characteristics, such as power, CPU, memory, and storage-related configurations.

Security research: Intel, in collaboration with Northeastern University, developed an end-to-end network slicing solution where each network slice could have different security properties and different service level agreements (SLAs). For example, one slice of the network could run the 5G UPF, or any other mobile core components in an Intel® SGX enclave. The extensions provided to the open mobile platform runtime operation controller (ROC) enable the creation of these slices and selection of specific slice properties. Northeastern University developed the necessary RAN open distributed unit (O-DU) functionality in the Open Air Interface radio while Intel developed a RAN slicing xApp following the latest O-RAN cell control and configuration (CCC) specification.

Future Research on the Open Platform Infrastructure

The open platform infrastructure is an ideal open source based infrastructure to research, develop, and validate advanced features in a real world environment. There is ongoing research in (i) energy efficiency to address challenges faced by mobile operators looking to optimize RAN and mobile energy to meet workload requirements, (ii) end-to-end AI/ML based resource allocation to deploy the minimal required resources to meet workload service level agreement, (iii) deployment and automation at scale, (iv) and end-to-end intent-driven orchestration automation.

Join us and the open source community in developing and deploying innovative edge services, innovating on the open platform infrastructure, and automating deployment and management of this platform at scale.