BESS UPF Deployment

This section describes how to configure and deploy BESS UPF.

Network Configuration

BESS UPF enabled edge setup requires three additional user plane subnets apart from the default K8S subnets.

  • enb: Used to provide eNB/gNB radios with connectivity to SD-Core and UPF.

  • access: Used to provide UPF with connectivity to eNBs.

  • core: Used to provide UPF with edge services, local breakout, and internet access. This naming comes from the P4 UPF, and is used for uniformity

To help your understanding, the following example ACE environment will be used in the rest of the guide.



Admin and out-of-band networks are not depicted in the diagram.





IP address


1 (gw:

management router





2 (gw:



3 (gw:

upf1 access


4 (gw:

management router

upf1 core

It is assumed that the management router has the only external routable address and acts as a router connecting the Aether pod to the outside.

This means that all uplink packets leaving the Aether pod need to be masqueraded with the external address of the management router or the k8smgmt address if the destination is Aether Central.

Also, in order for downlink traffic to UE to be delivered to its destination, it must be forwarded to the UPF’s core interface. This adds additional routes to the management router and L3 switch.


There is a work-in-progress on interface consolidation in BESS UPF, which will merge radio, access, and fabric into only one interface and simplify the configuration significantly.

Required Cluster Resources

Before proceeding with the deployment, make sure the cluster has sufficient resources to run BESS UPF by running the command below.

$ kubectl get nodes -o json | jq '.items[].status.allocatable'
  "cpu": "95",
  "ephemeral-storage": "1770223432846",
  "hugepages-1Gi": "32Gi",
  "": "32",
  "": "32",
  "memory": "360749956Ki",
  "pods": "110"

For best performance, BESS UPF requires the following resources:

  • 2 dedicated cores ("cpu")

  • 2 1GiB HugePages ("hugepages-1Gi")

  • 2 SRIOV Virtual Functions bound to vfio-pci driver ("")

For environments where these resources are not available, contact Ops team for advanced configuration.

Configure and Deploy

Download aether-app-configs if you don’t have it already in your development machine.

$ git clone "ssh://[username]"

Move the directory to apps/bess-upf/upf1 and create a Helm values file for the new cluster as shown below. Don’t forget to replace the IP addresses in the example configuration with the addresses of the actual cluster.

$ cd $WORKDIR/aether-app-configs/apps/bess-upf/upf1
$ mkdir overlays/prd-ace-test
$ vi overlays/prd-ace-test/values.yaml
# SPDX-FileCopyrightText: 2021-present Open Networking Foundation <>

      subnet: ""
      ip: ""
      gateway: ""
      vlan: 3
      ip: ""
      gateway: ""
      vlan: 4
  # Add below when connecting to 5G core
  #  upf.json:
  #    gtppsc: true
  #    cpiface:
  #      dnn: "internet" # should match the one configured in ROC

Update fleet.yaml in the same directory to let Fleet use the custom configuration when deploying BESS UPF to the new cluster.

$ vi fleet.yaml
# add following block at the end
- name: prd-ace-test
    matchLabels: ace-test
      - overlays/prd-ace-test/values.yaml

Submit your changes.

$ cd $WORKDIR/aether-app-configs
$ git status
$ git add .
$ git commit -m "Add BESS UPF configs for test ACE"
$ git review

Go to Fleet dashboard and wait until the cluster status becomes Active. It can take up to 1 min for Fleet to fetch the configuration updates.