Bandwidth-Allocated Network Slicing on 5G Core
BANS-5GC enables network slicing with specific bandwidth allocation policies on 5G core network by utilizing programmable data plane on transport network.
The 5GC uses free5GC, an open-source 5G core network.
The programmable data plane application for bandwidth management uses a research from 2019 20th APNOMS[1].
All functions are containerized and deployed on Kubernetes.
- Docker v18.06.2
- Kubernetes v1.16.2
- Calico v3.8
- Helm 3.0.0
- free5GC Stage 1 and Stage 2
- CPU 2 cores
- RAM 8GB
- NIC 2 cards (connecting to eNodeB and the Internet)
# On Kubernetes node
sudo apt install -y makeCurrently there are no gNodeB and UE for standalone 5GC available in the market yet. Procedure tests are provided for free5GC Stage 2. Skip this section if deploying Stage 2.
S1AP/SCTP protocol for 4GC is still remained between RAN and AMF in free5GC stage 1. Therefore, the following setup is required for field tests.
The setup assumes the eNodeB is directly connected to the server. Please modify eNodeB network configuration to work with your network environment.
# On Kubernetes node
# Configure network environment
sudo ip address add 192.168.3.2/24 dev ${ENODEB_INTF}
sudo sh -c 'echo 1 > /proc/sys/net/ipv4/ip_forward'Configure eNodeB settings
- IP address in subnet 192.168.3.0/24 (excluding ${ENODEB_INTF})
- Gateway/Router 192.168.3.2
- MME/AMF IP 192.168.2.2
NOTE 1: ${ENODEB_INTF} is server's interface connecting to eNodeB.
NOTE 2: For deploying 5GC only, it should be fine to set your desired address and subnet on both eNodeB and Kubernetes node interface connecting to eNodeB. Simply make sure packets can be correctly forwarded between two devices.
NOTE 3: For deploying 5GC with SDN-based transport and with BANS, ${ENODEB_INTF} should be any available address under subnet 192.168.3.0/24 by default. See Customizing Configuration section to customize the subnet in deployment.
NOTE 4: For deploying 5GC with SDN-based transport and with BANS, it's important to make sure the index of network interface connecting to the Internet is smaller than the one connecting to eNodeB.
Configure eNodeB settings
- IP address in subnet 192.168.3.0/24 (excluding 192.168.3.2 and 192.168.3.6)
- MME/AMF IP 192.168.3.2
NOTE 1: 192.168.3.2 and 192.168.3.6 are AMF's and UPF's IP address respectively by default. See Customizing Configuration section to customize two IP addresses.
# Deploy
make
# or
make bans-5gcv1A OvS-based Mininet pod is placed between eNodeB and UPF pod.
Open vSwitches are controlled by ONOS.
# Deploy
make bans-5gcv1-ovs# Deploy
make bans-5gcv1-bmv2
# Override default slice configuration
SLICE_CONFIG=/path/to/file make bans-5gcv1-bmv2
# Add new onos slice on existing deployment
SLICE_CONFIG=/path/to/file make onos-bw-sliceExample slice configuration is placed at deploy/slice.json.
# Deploy
# The argument `SRIOV_INTF` is required for server setup and creating SR-IOV resources
SRIOV_INTF=devicename make bans-5gcv1-sriovNOTE 1: The make script performs server setup for SR-IOV, such as loading device's kernel module and creating required virtual functions, which is experimental and only be tested with Intel Ethernet adapters. Manually configure your server for SR-IOV devices if make fails at target sriov-server-setup.
# Deploy
make bans-5gcv2# Enter RANSIM pod
export RANSIM_POD=$( kubectl get pod -l app.kubernetes.io/instance=free5gc -l app.kubernetes.io/name=ransim -o jsonpath='{.items[0].metadata.name}' )
kubectl exec -ti $RANSIM_POD bash
# Run procedure test in RANSIM pod
# If deploying with network slicing disabled
cd src/test
go test -v -vet=off -run TestRegistration
# If deploying with network slicing enabled
# Replace <ue-id> with slice ID in configuration .global.slices
cd src/test
go test -v -vet=off -run TestRegistration -ue-id=<ue-id>NOTE 1: For deploying with network slicing enabled, the Registration procedure tests should be executed in the incremental order of <ue-id>, since NGAP IDs in registration are expected to be incremented strictly.
# Customize configuration
BANSVALUES=/path/to/file make [target]Default configuration files for deployment locate in helm-charts/configs/<target>.yaml.
See helm-charts/configs/README.md for more informations.
Web UI to subscriber database is running on port 30300. Default username and password are admin and 1423.
ONOS GUI is exposed on port 30181. Default username and password are onos and rocks.
Description
Especially in deploying 5GC with SDN-based transport and with BANS, ARP cache of UPF pod is incompleted on the Kubernetes node.
Root Cause
Calico CNI exploits routing table and iptables infrastructure for communication of workloads, and Kernel sets source IP address of ARP requests for workloads to address of the first available network interface in the order of interface index numbers, except loopback addresses.
The problem takes place when network interface connecting to eNodeB has the smallest index number among all interfaces with available addresses, which means it takes priority over others on the decision of source address of ARP requests. Since UPF pod has a route to direct traffic with the destination of eNodeB to Mininet in such two deployments, replies to ARP requests from UPF cannot be correctly sent back to Kubernetes node.
Workaround
Add a dummy address, other than loopback addresses, on the loopback interface to force Kernel set source address of ARP requests to it, since loopback interface has the smallest index number.
Please see stevenchiu30801/bans-5gc-images.
Note that ONOS image with the bandwidth management application and free5GC image of unreleased version are kept private. Please checkout branch release-1.2 for public release v2.0.2 of free5GC deployment.
[1] P4-Enabled Bandwidth Management
[2] opencord/automation-tools
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent