5G Private Wireless Experience Kit with Integrated RAN
Intel® Smart Edge Open experience kits provide customized infrastructure deployments for common network and on-premises edge use cases. Combining Intel cloud-native technologies, wireless networking, and high-performance compute, experience kits let you deliver AI, video, and other services optimized for performance at the edge.
The 5G Private Wireless Experience Kit with Integrated RAN accelerates the development of edge platforms and architectures for deploying private networks.
5G Private Wireless Experience Kit with Integrated RAN
The assumed 3GPP deployment architecture is based on the figure below from 3GPP 23.501 Rel15 which shows the reference point representation for concurrent access to two (e.g. local and central) data networks (single PDU Session option).
5G Private Wireless with Integrated RAN deploys the following building blocks:
NG-RAN (gNodeB DU and CU)
User plane function (UPF)
Data network (edge applications)
5GC control plane functions including the access and mobility management function (AMF) and session management function (SMF)
These are highlighted in yellow in the diagram below.
Figure 1 - 3GPP Network
NOTE: The reference solution assumes use of our partner Radisys’s simplified 5G core network, so only the AMF and SMF functions are highlighted here. The deployment can be extended to support other control plane functions as well.
Please contact Radisys to get 5G core network functions, or contact your local Intel® representative for more information.
The 5G Private Wireless Experience Kit with Integrated RAN supports a single orchestration domain, optimizing the edge node to support both network functions, such as DU/CU, AMF, SMF, and UPF, and applications.
The 5G Private Wireless Experience Kit with Integrated RAN includes building blocks for 5G DU/CU and UPF functionality, and for running applications and their associated hardware accelerators. The below diagram shows the logical deployment with the Intel® Smart Edge Open Building Blocks.
Figure 2 - Building blocks for 5G Private Wireless with Integrated RAN
Figure 3 - 5G Private Wireless with Integrated RAN Architecture
The 5G Private Wireless Experience Kit with Integrated RAN supports 2 Intel® Xeon® Scalable Processor nodes in a single cluster, one serving as an edge node and the second as a Kubernetes control plane node.
The UPF is deployed using SR-IOV-Device plugin and SRIOV-CNI allowing direct access to the network interfaces used for connection to the CU and back haul.
For high throughput workloads such as UPF network function, it is recommended to use single root input/output (SR-IOV) pass-through the physical function (PF) or the virtual function (VF), as required.
The 5G Private Wireless Experience Kit with Integrated RAN leverages the simple switching capability in the NIC can be used to send traffic from one application to another, as there is a direct path of communication required between the UPF and the Data plane.
The applications are deployed on the same edge node as the UPF and CU/DU.
The following building blocks are supported in Intel® Smart Edge Open:
QAT and SR-IOV Device plugin for Kubernetes
Node Feature Discovery (NFD): Software that detects hardware features available on each node in a Kubernetes cluster and advertises those features using node labels.
Topology Manager: Allows users to align their CPU and peripheral device allocations by NUMA node.
Kubevirt: Supports running legacy applications in virtual machine (VM) mode and allows SR-IOV Ethernet interfaces to be allocated to VMs.
Precision Time Protocol (PTP): Provides time synchronization between machines connected through Ethernet. The primary clock serves as a reference clock for secondary nodes. A grand master clock (GMC) can be used to precisely set the primary clock.
The 5G Private Wireless Experience Kit with Integrated RAN hosts the 5G access network and core network functions on a single cluster.
NOTE: Experience kits don’t include 5G binaries. If you require binaries, Intel can refer you to a partner who can provide them. Contact your local Intel representative for more information.
Figure 4 - Cluster provisioning in the 5G Private Wireless Experience Kit with Integrated RAN
NOTE: You will need privileges to deploy and run the CU/DU for the relevant software version used in the reference architecture. Contact Radisys or your local Intel representative for more information.
CPU Cores Allocation
5G network functions such as DU and UPF are real-time, time-sensitive applications that require the allocation of dedicated CPU cores.
The example below shows CPU core allocation using the default CPU configuration with hyper-threading enabled and CPU MADT core enumeration set to linear. See an example configuration below: Set isolated CPU cores
Table 1 - 5G CPU allocation for the Private Wireless Experience Kit with Integrated RAN
Supported Edge Applications
The 5G Private Wireless Experience Kit with Integrated RAN uses an architectural paradigm that enables convergence of edge services and applications across different market segments. This is demonstrated by taking diverse workloads native to different segments and successfully integrating within a common platform. As a reference, consider the following applications from different market segments.
Smart city: Capture of live camera streams to monitor and measure pedestrian and vehicle movement within a zone.
Industrial: Monitoring of the manufacturing quality of an industrial line, the capture of video streams focuses on manufactured devices on an assembly line and the real-time removal of identified defect parts.
While these use cases are addressing different market segments, they both have similar requirements:
Capture video either from a live stream from a camera, or streamed from a recorded file.
Process that video using inference with a trained machine learning model, computer vision filters, etc.
Trigger business control logic based on the results of the video processing.
Video processing is inherently compute intensive and, in most cases, especially in edge processing, video processing becomes the bottleneck in user applications. This, ultimately, impacts service KPIs such as frames-per-second, number of parallel streams, latency, etc.
Therefore, pre-trained models, performing numerical precision conversions, offloading to video accelerators, heterogeneous processing and asynchronous execution across multiple types of processors all of which increase video throughput are extremely vital in edge video processing. However these requirements can significantly complicate software development, requiring expertise that is rare in engineering teams and increasing the time-to-market.
The 5G Private Wireless Experience Kit with Integrated RAN offers a sample edge application based on OpenVINO (Open Visual Inference and Neural Network Optimization). The Intel® Distribution of OpenVINO™ toolkit helps developers and data scientists speed up computer vision workloads, streamline deep learning inference and deployments, and enable easy, heterogeneous execution across Intel® architecture platforms from edge to cloud. It helps to unleash deep learning inference using a common API, streamlining deep learning inference and deployment using standard or custom layers without the overhead of frameworks.
The 5G Private Wireless with Integrated RAN Experience Kit is designed to run on standard, off-the-shelf servers with with 3rd Generation Intel® Xeon® Scalable Processors. The experience kit has been validated to run on a Dell Server R750. Please refer to Smart Edge Open 21.09 release notes for more detailed Dell R750 configurations.
Hardware accelerators can be used to increase the performance of certain workloads. Use the Intel® Smart Edge Open Kubernetes control plane node to assign accelerators to a specific container whose workload you are targeting.
The Intel® QuickAssist Adapter provides customers with a scalable, flexible, and extendable way to offer Intel® QuickAssist Technology (Intel® QAT) crypto acceleration and compression capabilities to their existing product lines. Intel® QuickAssist Technology provides hardware acceleration to assist with the performance demands of securing and routing Internet traffic and other workloads, such as compression and wireless 4G LTE and 5G gNB(g-NodeB) algorithm offload, thereby reserving processor cycles for application and control processing.
Intel® ACC100 eASIC
The Intel® vRAN Dedicated Accelerator ACC100 Adapter accelerates 4G and 5G virtualized radio access network (vRAN) workloads, which in turn increases the overall compute capacity of commercial, off-the-shelf platforms.
Reduced platform power, E2E latency and Intel® CPU core count requirements as well as increase in cell capacity than existing programmable accelerator.
Accelerates both 4G and 5G data concurrently.
Lowers development cost using commercial off the shelf (COTS) servers.
Intel® FPGA(Field Programmable Gate Array) Programmable Acceleration Card N3000 (Intel® FPGA PAC N3000)
The Intel® FPGA PAC plays a key role in accelerating certain types of workloads, which in turn increases the overall compute capacity of a commercial, off-the-shelf platform. FPGA benefits include:
Flexibility - FPGA functionality can change upon every power up of the device.
Acceleration - Increase your system performance by offload workload from CPU to FPGA.
Integration - Modern FPGAs include on-die processors, transceiver I/Os at 28 Gbps (or faster), RAM blocks, DSP engines, and more.
Total Cost of Ownership (TCO) - While ASICs may cost less per unit than an equivalent FPGA, building them requires a non-recurring expense (NRE), expensive software tools, specialized design teams, and long manufacturing cycles. FPGA can help to get products to market faster.
The Intel® FPGA PAC N3000 is a full-duplex, 100 Gbps in-system, re-programmable acceleration card for multi-workload networking application acceleration. It has an optimal memory mixture designed for network functions, with an integrated network interface card (NIC) in a small form factor that enables high throughput, low latency, and low power per bit for a custom networking pipeline.
Please contact your local Intel® representative for more information about mobile phone and USIM card.
Table 2 - 5G Private Wireless with Integrated RAN Hardware List
NOTE: For server, suggest that number of CPU cores is not less than 26, memory size is not less than 100G, and the disk capacity is not less than 300G.
Overview of Network Functions
Verified FlexRAN BBU v20.11.
Verified Radisys L2DU v2.2.
Verified Radisys L2L3CU v2.2.
UPF,AMF and SMF
Verified Radisys 5GC v2.2.
Table 3 - 5G Private Wireless with Integrated RAN software list
NOTE: Please contact your local Intel® representative for more information about 5G network functions.
Install CentOS 7.9 Minimal on both the edge node and the controller.
The BIOS settings on the edge node must be properly set in order for the Intel® Smart Edge Open building blocks to function correctly. They may be set either during deployment of the experience kit, or manually. The settings that must be set are:
Enable Intel® Hyper-Threading Technology
Enable Intel® Virtualization Technology
Enable Intel® Virtualization Technology for Directed I/O
Enable SR-IOV Support
Set up the Target Platform
Perform the following steps on the target machine before deployment:
Ensure the target system gets its IP address automatically on boot.
Example command to check IP addresses: hostname -I
Change the target system’s hostname.
Edit the /etc/hostname file:
* vi /etc/hostname
* Press the Insert key to enter Insert mode
* Delete the old hostname and replace it with the new one. The new hostname can be any combination of valid characters, preferably a meaningful name.
* Exit the vi editor by pressing the Esc key, then typing :wq and pressing the Enter key.
Edit the /etc/hosts file:
* vi /etc/hosts
* Press the Insert key to enter Insert mode
* Add a space at the end of both lines in the file and write hostname after it.
* Exit the vi editor by pressing the Esc key, then typing :wq and pressing the Enter key.
Reboot the target machine.
Steps to be performed
These steps are performed on the machine where the Ansible playbook will be run.
Copy the SSH key from the machine where the Ansible playbook is going to be run, to the target machine. Example commands:
NOTE: Generate ssh key if is not present on the machine: ssh-keygen -t rsa (Press enter key to apply default values)
Repeat for each target machine.
NOTE: Replace TARGET_IP with the actual IP address of the target machine.
Clone the ido-converged-edge-experience-kits repo from github.com/smart-edge-open using a Git token.
Provide the target machine’s IP addresses for 5G Private Wireless with Integrated RAN deployment in ido-converged-edge-experience-kits/inventory.yml. For the cluster setup, set the IP addresses for controller and node01. In the same file, define the details for 5G Private Wireless with Integrated RAN cluster deployment.
Edit ido-converged-edge-experience-kits/inventory/default/group_vars/all/10-open.yml to configure settings for deployment.
Proxy, if required.
# Setup proxy on the machine - required if the Internet is accessible via proxyproxy_enable:true# Clear previous proxy settingsproxy_remove_old:true# Proxy URLs to be used for HTTP, HTTPS and FTPproxy_http:"http://proxy.example.org:3128"proxy_https:"http://proxy.example.org:3129"proxy_ftp:"http://proxy.example.org:3129"# Proxy to be used by YUM (/etc/yum.conf)proxy_yum:""# No proxy setting contains addresses and networks that should not be accessed using proxy (e.g. local network, Kubernetes CNI networks)proxy_noproxy:"127.0.0.1,localhost,192.168.1.0/24"
# CPUs to be isolated (for RT procesess)cpu_isol:"1-31,33-63,65-95,97-127"# CPUs not to be isolate (for non-RT processes) - minimum of two OS cores necessary for controllercpu_os:"0,32,64,96"
Configure the PTP connection with the PTP Grand Master:
ptp_port:"p5p3"# Interface name for PTP single node setupptp_port_gm:"p5p3"# Interface logical name (PF) used for PTP connectionptp_network_transport:"-4"# UDP IPv4 network transport.ptp_port_ip:"192.168.1.160"# Static IP for the server port connected to Grand Masterptp_port_cidr:"24"gm_ip:"192.168.1.170"# Grand Master IP.
Set up hugepages settings
# Size of a single hugepage (2M or 1G)hugepage_size:"1G"# Amount of hugepageshugepage_amount:"60"
Provide Intel® Save and Restore System Configuration Utility (SYSCFG) package:
The SYSCFG package must be downloaded and stored inside Converged Edge Experience Kits biosfw/ directory as syscfg_package.zip: ido-converged-edge-experience-kits/ceek/biosfw/syscfg_package.zip
For more details about BIOS-FW feature please refer to openness-bios.
Provide OPAE package ( Optional step with FPGA N3000 ):
Place OPAE_SDK_1.3.7-5_el7.zip inside ido-converged-edge-experience-kits/ceek/opae_fpga directory. The package can be obtained as part of Intel® FPGA PAC N3000 OPAE beta release. To obtain the package, contact your Intel representative.
For more details about FPGA support please refer to openness-fpga.
Copy gNodeB and 5GCN images and deployment files
The binaries from gNodeB and 5GCN are required to be copied from the deployment machine so it needs to have atleast 50-60 GB of storage available.
2. Set the upload path for `GNodeB` prerequisites.
Edit the `ido-converged-edge-experience-kits/flavors/pwek-all-in-one/all.yml` file to customize the upload path.
# on edge node
Configure PTP Time Synchronization
Synchronize the edge node using PTP to allow it to connect with the RRH (remote radio head).
Not every NIC supports hardware time stamping. To verify if the NIC supports hardware time stamping, run the ethtool command for the interface in use.
ethtool -T eno3
Time stamping parameters for eno3:
PTP Hardware Clock: 0
Hardware Transmit Timestamp Modes:
Hardware Receive Filter Modes:
For software time stamping support, the parameters list should include:
For hardware time stamping support, the parameters list should include:
The GMC must be properly configured and connected to the server’s ETH port.
Edit the ido-converged-edge-experience-kits/inventory.yml file.
Add node01 should to the ptp_slave_group.
Comment out any content under ptp_master.
Enable server synchronization in the ido-converged-edge-experience-kits/flavors/pwek-all-in-one/edgenode_group.yml file.
If a GMC is connected and the node server should be synchronized, edit the ido-converged-edge-experience-kits/flavors/pwek-all-in-one/edgenode_group.yml file.
For single-node setup, which is the default mode for the experience kit, ptp_port keeps the host’s interface connected to the grand master clock, e.g.:
The ptp_network_transportvariable keeps network transport for ptp. Choose "-4" for default CERA setup. The gm_ip variable should contain the GMC’s IP address. The Ansible scripts set the IP on the interface connected to the GMC, according to the values in the variables ptp_port_ip and ptp_port_cidr.
# Valid options:# -2 Select the IEEE 802.3 network transport.# -4 Select the UDP IPv4 network transport.ptp_network_transport:"-4"# Grand Master IP, e.g.:# gm_ip: "169.254.99.9"gm_ip:"169.254.99.9"# - ptp_port_ip contains a static IP for the server port connected to GMC, e.g.:# ptp_port_ip: "169.254.99.175"# - ptp_port_cidr - CIDR for IP from, e.g.:# ptp_port_cidr: "24"ptp_port_ip:"169.254.99.175"ptp_port_cidr:"24"
Port State: Master
Delay Mechanism: E2E
Network Protocol: IPv4
Sync Interval: 0
Delay Request Interval: 0
Pdelay Request Interval: 0
Announce Interval: 3
Announce Receipt Timeout: 3
Multicast/Unicast Operation: Unicast
Profile: Default (1588 v2)
Two step clock: FALSE
Clock class: 248
Clock accuracy: 254
Offset scaled log: 65535
Priority 1: 128
Priority 2: 128
Domain number: 0
Slave only: FALSE
5G Core Network Functions
This section describes in detail how to build 5G core network function images and configure Ansible for deployment.
The User Plane Function (UPF) is the part of the 5G core network responsible for routing packets. It has two separate interfaces for N4/N9 and N6 data lines. The N4/N9 interface is used for connection with dUPF and AMF/SMF (locally). The N6 interface is used for connection with EDGE-APP, dUPF and Remote-DN (locally).
During deployment of the edge cluster, the UPF component is automatically deployed to the edge node as a pod.
To deploy UPF correctly you must provide a Docker image to Docker Repository on target nodes. Ansible only supports import of UPF and OAM-UPF docker images released by Radisys.
PWEK 5GC path in file ido-converged-edge-experience-kits/roles/applications/pwek_5gc/defaults/main.yml used for docker images, shell scripts and helm-charts.
# Paths for storing pwek related resourcesremote_pwek_path:"/opt/pwek"pwek_charts_path:"/charts/"pwek_images_path:"/images/"pwek_scripts_path:"/scripts/
The UPF is configured automatically during the deployment.
Deploy the Access and Mobility Management Function (AMF) and Session Management Function (SMF)
AMF-SMF are the 5G core architecture functions responsible for access and mobility management, and session management respectively. Together, the AMF-SMF establishes sessions and manages data plane packages.
The AMF-SMF component is deployed on the PWEK Controller node. It communicates with the UPF and dUPF, which must be deployed and configured before this step.
AMF and SMF Docker images must be provided to the Docker repository on the target controller node.
NOTE: Contact Radisys get AMF and SMF Docker images.
The AMF-SMF is configured automatically during the deployment.
Onboard the Intel® Distribution of OpenVINO™ toolkit to the Edge Node
This section describes how to onboard the Intel® Distribution of OpenVINO™ toolkit to the edge node.
Get the IP address of the SR-IOV interface attached to OpenVINO consumer pod by running the ip a command in the consumer pod. The ip address is in 220.127.116.11/24 subnet.
Select the “Open Network Stream” option.
Input the address “rtmp://:5000/live/out.flv". Here, the openvino_consumer_pod_sriov_ip is the address get by `ip a`.
The video will load after one second.
This guide walked you through deploying the 5G Private Wireless Experience Kit with Integrated RAN. The reference implementation of Intel® Smart Edge Open created by this process can be used to efficiently deploy, manage, and optimize network functions and applications specifically for an on-premises private wireless network. You can continue to customize this deployment to meet your own use cases.