1 of 15

2 of 15

Vitumbiko Mafeni | Jangwon Lee

Making IoT Simple

Leveraging Akri & Nephio in a Hybrid Cloud Environment

3 of 15

Vitumbiko Mafeni�Research Engineer�SSU IISTRC

Jangwon Lee�Research Engineer�SSU IISTRC

Who we are

November 12, 2024�Salt Lake City

4 of 15

5 of 15

What we have on the table: Nephio

For the purposes of Nephio deployment, two categories of Nephio:

Management cluster: This is where the majority of Nephio components are installed. It is dedicated to manage the deployment and lifecycle management of network functions that will be deployed on workload clusters.
Workload cluster: This is where the actual network function workloads are deployed and running.

Workload Clusters

Functionalities

Create Kubernetes clusters. This functionality is based on cluster API. At this time only KIND clusters creation is supported.
Fully automated deployment of UPF, SMF and AMF services of free5Gc . These are deployed on multiple clusters based on user’s intent expressed via CRDs.
Inter cluster networking setup - containerlab.
Deployment of other free5gc functions. Some manual configuration such as IP addresses may be needed for these services.

source: https://docs.nephio.org/docs/

6 of 15

What we have on the table: Akri

There are two Akri CRDs:

Configuration: Akri users will create Configurations to describe what resources should be discovered and what pod should be deployed on the nodes that discover a resource.
Instance: Each Instance represents an individual resource that is visible to the cluster. So, if there are 5 IP cameras visible to the cluster, there will be 5 Instances.

Extending Kubernetes to the Edge with Akri

Edge environment includes various sensors, controllers, and MCU class devices producing data and performing actions.
Kubernetes clusters need to easily utilize heterogeneous “leaf devices” at the edge.
Many leaf devices are too small to run Kubernetes natively.
Akri: Open-source project to expose leaf devices as resources in Kubernetes clusters.
Akri extends Kubernetes' device plugin framework, originally designed for cloud hardware like GPUs.
Akri adapts the framework for edge devices with unique communication protocols and intermittent availability.

source: https://docs.akri.sh/

7 of 15

The Challenge in Nephio & Akri

Automating IoT infrastructure provisioning.
Deploying and configuring IoT applications across multiple cloud providers.
Managing thousands of IoT devices across different geographic locations.

We can solve these challenges with Nephio and Akri

When it comes to IoT, many engineers encounter challenges in automating infrastructure provisioning, deploying IoT applications across multiple sites, and managing thousands of IoT devices spread across different geographic locations. The complexity and dynamic nature of IoT ecosystems, characterized by diverse protocols and ever-changing statuses, makes it overwhelming.��Imagine an automated approach that seamlessly deploys and configures these devices in real-time. GitOps has emerged as a leading trend to facilitate collaboration, configuration, and deployment of intricate and dynamic applications. Join us for this session where we present our solution for deploying dynamic IoT applications over a Private 5G network in a hybrid cloud environment. This session will give details on how the integration of CNCF projects such as Nephio, and Akri can transform your IoT management through the adoption of GitOps principles.

8 of 15

Solution Overview

9 of 15

Solution 1 – Nephio AWS deployment PoC

�After studying and attending Nephio community meetings, we have gained insights into, experimented and resolved common issues encountered in the Nephio Project.��Firstly, we have successfully addressed some of the open issues in the current release of Nephio. Specifically, unlike Nephio R2, which only demonstrates proof of concepts on Kind clusters, our hybrid cloud setup utilizes real self-managed Kubernetes clusters on AWS and OpenStack.��Secondly, Nephio R2's current release focuses solely on provisioning and managing 5G core network functions within a multi-cluster setup limited to a single cloud provider at a time (such as Kind clusters or Google Cloud). To overcome this limitation, we opted to slice the network functions. As a result, the User Plane Function (UPF) and 5G core are interconnected and situated on AWS, while the Radio Access Network (RAN) component resides on a private cloud (OpenStack).��

10 of 15

Solution 2 – Nephio Akri Integration – IoT PoC

Akri Components running on Nephio Workload clusters, inferencing intensive services running on Nephio Workload cluster on AWS

11 of 15

Nephio Package Management

Helm currently not supported in the Nephio UI*

New Registered Akri package

IoT cluster creation on OpenStack with Akri Packages

12 of 15

Clusters Created

Workload Cluster

Created Workload Cluster -> kubectl

13 of 15

Argo Application helm configuration

ArgoCD Configs + Syncing Akri Components

Repo

Akri Components

Cluster repo/Nephio cluster name/Argo App

14 of 15

Akri Components Over WorkloadCluster

Discovering & Using USB Cameras

Akri Helm Values in a git repo configurations deployed through ArgoCD and managed through Nephio Management cluster

15 of 15

Takeaway

Multi-Site IoT Challenges
Akri Integration
Nephio and Hybrid Cloud

SSU-IISTRC

Github

Contact Info

Mafeni Vitumbiko: vitumafeni@dcn.ssu.ac.kr

Jangwon Lee: jangwon.lee@dcn.ssu.ac.kr

Multi-Site IoT Challenges: Engineers face difficulties automating infrastructure, deploying IoT applications across multiple sites, and managing diverse devices spread across geographies. GitOps simplifies IoT management by automating collaboration, configuration, and deployment of complex, dynamic applications and devices in Akri.
Akri Integration: Akri was integrated to manage IoT devices, with custom IoT packages deployed through ArgoCD managed by Nephio for package orchestration, improving scalability and efficiency across hybrid cloud environments.
Nephio and Hybrid Cloud: The solution enhances Nephio R2 by using self-managed Kubernetes clusters across AWS and OpenStack, overcoming single-cloud limitations and enabling hybrid cloud IoT management. At the same time, showing a proof of concept of nephio in real cloud rather than Kind clusters