Copyright (C) 2019-2023 Intel Corporation
FlexRAN is Intel 4G and 5G baseband PHY Reference Design, which uses Xeon® series Processor with Intel Architecture. This 5GNR Reference PHY consists of a L1 binary which are validated on a Intel® Xeon® SkyLake / CascadeLake platforms and demonstrates the capabilities of the software running different 5GNR L1 features. It implemented the relevant functions in [3GPP TS 38.211, 212, 213, 214 and 215].
The L1 application itself support private interface, togther with the translate layer application, FAPI interface was supported. FAPI was defined in small cell forum, you can download through https://www.smallcellforum.org/. the FAPI version 222.10.02 was adopted in current release. Details of FAPI and pravite API mapping, you could refer the FAPI document in O-RAN software community O-DU low document https://gerrit.o-ran-sc.org/r/o-du/phy. Regard to FEC functionalities, there are two options supported by the application
- FEC running in IA with DPDK BBDev associated with specific SW FEC SDKs
- FEC running in FPGA/ASIC with DPDK BBDev interface and associate driver
Details install guide can be find in below installation section
The L1 PHY Application can be tested for functionality and performance (meeting realtime runtime budgets) using an application called the testmac. These are basically testing complete API to IQ samples of all modules within the L1 application. The main purpose of these 2 applications are:
- To generate APIs to the PHY and receive APIs from the PHY.
- At the beginning of a UL test, they provide reference IQ samples to the PHY and at the end of a DL test, they receive the IQ samples from the PHY. There are multiple test folders which have:
- Config files for multi-slot (xml format)
- Reference UL IQ Samples to use as Input to PHY (for UL test)
- Reference DL IQ Samples to use as reference Output to compare from PHY (for DL test)
- Reference UL Payload files (for UL Test) which have all the payloads for PUSCH / PUCCH / PRACH / SRS / others At the end of the test, both tools compare the reference DL IQ / UL Payloads and generate a pass / fail report. These tools are explained in more detail in the next few sections. In terms of testing we have 3 different types of tests: DL: Downlink Only UL: Uplink Only FD: Full Duplex - DL and UL Each test type has a bunch of test cases (with unique testNum) and these are defined in the Bit Exact Test Cases section
To run the L1 PHY application, you should configure the BIOS and OS of the server. Take Intel Cascade Lake SP server as an example here.
Advanced -> Power & Performance
CPU Power and Performance Policy : Performance
Advanced -> Power & Performance -> Uncore Power Management
Uncore Frequency Scaling : Disabled Performance P-limit : Disabled
Advanced -> Power -> CPU P State Control
Enhanced Intel SpeedStep (R) Tech : Enabled Intel Configurable TDP : Enabled Configurable TDP Level : Level 2 Intel(R) Turbo Boost Technology : Enabled Energey Efficient Turbo : Disabled
Advanced -> Power & Performance -> Hardware P States
Hardware P-States : Disabled
Advanced -> Power & Performance -> CPU C States Control
Package C-State : C0/C1 state C1E : Disabled Processor C6 : Disabled
First you should install CentOS 7 (7.5+) from CentOS official site, then install Real Time kernel and required tools to tune the OS real timer performance. CentOS 7.7 and RT kernel kernel-rt-3.10.0-1062.12.1.rt56.1042 have been validated for this release.
yum install -y kernel-rt kernel-rt-devel kernel-rt-kvm rtctl rt-setup rt-tests tuna tuned-profiles-nfv tuned-profiles-nfv-host tuned-profiles-nfv-guest qemu-kvm-tools-ev
Below give one example to tune the RT kernel performance for Intel Cascade Lake CPU 6248 Edit /etc/tuned/realtime-virtual-host-variables.conf to add isolated_cores=1-19:
# Examples:
# isolated_cores=2,4-7
# isolated_cores=2-23
isolated_cores=1-19
Core 1-19 are isolated from the host OS and dedicated for real time tasks.
Run below command to activate Real Time Profile
tuned-adm profile realtime-virtual-host
Configure kernel command line Edit /etc/default/grub and append the following to the GRUB_CMDLINE_LINUX:
"processor.max_cstate=1 intel_idle.max_cstate=0 intel_pstate=disable idle=poll default_hugepagesz=1G hugepagesz=1G hugepages=32 intel_iommu=on iommu=pt selinux=0 enforcing=0 nmi_watchdog=0 audit=0 mce=off kthread_cpus=0 irqaffinity=0"
Add the following:
GRUB_CMDLINE_LINUX_DEFAULT="${GRUB_CMDLINE_LINUX_DEFAULT:+$GRUB_CMDLINE_LINUX_DEFAULT }\$tuned_params"GRUB_INITRD_OVERLAY="${GRUB_INITRD_OVERLAY:+$GRUB_INITRD_OVERLAY }\$tuned_initrd"
After the change, the grub file runs the following command to update the grub:
grub2-mkconfig -o /boot/grub2/grub.cfg
Reboot the server, and check the kernel parameter, which should look like:
cat /proc/cmdline
BOOT_IMAGE=/vmlinuz-3.10.0-1062.12.1.rt56.1042.el7.x86_64 root=UUID=9b3e69f6-88af-4af1-8964-238879b4f282 ro crashkernel=auto rd.lvm.lv=centos/root rd.lvm.lv=centos/swap rhgb quiet processor.max_cstate=1 intel_idle.max_cstate=0 intel_pstate=disable idle=poll default_hugepagesz=1G hugepagesz=1G hugepages=32 intel_iommu=on iommu=pt selinux=0 enforcing=0 nmi_watchdog=0 audit=0 mce=off kthread_cpus=0 irqaffinity=0 skew_tick=1 isolcpus=1-19 intel_pstate=disable nosoftlockup nohz=on nohz_full=1-19 rcu_nocbs=1-19
Set CPU frequency using msr-tools
git clone https://github.com/intel/msr-tools/
cd msr-tools/
git checkout msr-tools-1.3
make
cat <<EOF > turbo-2.5G.sh
#!/bin/sh
for i in {0..39}
do
#Set core 0-39 to 2.5GHz (0x1900). Please change according to your CPU model
./wrmsr -p \${i} 0x199 0x1900
done
#Set Uncore to Max
./wrmsr -p 0 0x620 0x1e1e
./wrmsr -p 39 0x620 0x1e1e
EOF
chmod 755 turbo-2.5G.sh
sh turbo-2.5G.sh
Set CPU Frequency Policy to Performance.
cpupower frequency-set -g performance
RT Test and Verify the test result
cyclictest -m -n -p95 -d0 -a 1-16 -t 16
/dev/cpu_dma_latency set to 0us
policy: fifo: loadavg: 0.00 0.01 0.05 1/702 25564
T: 0 (25549) P:95 I:1000 C: 5796 Min: 4 Act: 5 Avg: 4 Max: 6
T: 1 (25550) P:95 I:1000 C: 5797 Min: 4 Act: 5 Avg: 4 Max: 6
T: 2 (25551) P:95 I:1000 C: 5791 Min: 4 Act: 5 Avg: 4 Max: 6
T: 3 (25552) P:95 I:1000 C: 5788 Min: 4 Act: 4 Avg: 4 Max: 6
T: 4 (25553) P:95 I:1000 C: 5785 Min: 4 Act: 4 Avg: 4 Max: 6
T: 5 (25554) P:95 I:1000 C: 5782 Min: 4 Act: 5 Avg: 4 Max: 6
T: 6 (25555) P:95 I:1000 C: 5778 Min: 4 Act: 5 Avg: 4 Max: 6
T: 7 (25556) P:95 I:1000 C: 5775 Min: 4 Act: 5 Avg: 4 Max: 6
T: 8 (25557) P:95 I:1000 C: 5772 Min: 4 Act: 5 Avg: 4 Max: 6
T: 9 (25558) P:95 I:1000 C: 5768 Min: 4 Act: 5 Avg: 4 Max: 6
T:10 (25559) P:95 I:1000 C: 5765 Min: 4 Act: 5 Avg: 4 Max: 6
T:11 (25560) P:95 I:1000 C: 5762 Min: 4 Act: 5 Avg: 4 Max: 6
T:12 (25561) P:95 I:1000 C: 5758 Min: 5 Act: 5 Avg: 5 Max: 5
T:13 (25562) P:95 I:1000 C: 5758 Min: 4 Act: 5 Avg: 4 Max: 5
T:14 (25563) P:95 I:1000 C: 5758 Min: 4 Act: 5 Avg: 4 Max: 5
T:15 (25564) P:95 I:1000 C: 5758 Min: 4 Act: 5 Avg: 4 Max: 5
Run at least 12/24 hours for rigid performance validation. For quick performance validation, 15 minutes is recommended. Pay attention to the Avg. and Max. On a well-tuned platform, the numbers should be similar.
Follow the steps in O-RAN SC O-DU Low project document to build WLS, FAPI TM and FH lib. Get through link https://docs.o-ran-sc.org/projects/o-ran-sc-o-du-phy/en/latest/index.html
Follow the steps in O-RAN SC O-DU Low project document to run L1 and associate testmac testing. Get through link https://docs.o-ran-sc.org/projects/o-ran-sc-o-du-phy/en/latest/index.html
Build O-DU docker image (build-oran-l1-image.sh will be updated to support the OneAPI compiler which is preferred over icc)
Please follow the build procedures on baremetal host to build FAPI/WLS/FH and copy the code/binaries to the oran_release folder on your build server. The docker build will just copy the pre-built binaries to docker image.
oran_release/phy
oran_release/FlexRAN
Set the network proxy and ORAN_RELEASE environment variable before build
export http_proxy=<your_proxy>
export https_proxy=<your_proxy>
export no_proxy="localhost,127.0.0.1"
export ORAN_RELEASE=<TOP_FOLDER>/oran_release
Update ${ORAN_RELEASE}/phy/setupenv.sh, change DIR_ROOT to /opt/oran_release. Run below script to trigger the docker build:
sh ${ORAN_RELEASE}/FlexRAN/docker/build-oran-l1-image.sh
After the docker build finishes, docker image with oran-release tag will be generated
…
Successfully tagged oran-release-du:4.0
You can push the O-DU docker image to your local docker registry where Kubernetes cluster can pull the image. Or you can copy the image locally to your Kubernetes worker node. In your worker node, make sure you have configured the system/tools correctly following previous sections for running L1 in baremetal host.
- Assume intel system studio is installed in /opt/intel/system_studio_2019/
- Make sure you have enough 1G hugepages
Update the Kubernetes pod yaml configuration for your system configuration
cat ${ORAN_RELEASE}/FlexRAN/docker/oran-release-du.yaml
apiVersion: v1
kind: Pod
metadata:
labels:
app: oran-release-du-pod
name: oran-release-du-pod
spec:
containers:
- securityContext:
privileged: false
capabilities:
add:
- SYS_ADMIN
- IPC_LOCK
- SYS_NICE
command:
- sleep
- infinity
tty: true
stdin: true
image: oran-release-du:4.0
name: l1app
resources:
requests:
memory: "16Gi"
hugepages-1Gi: 2Gi
limits:
memory: "16Gi"
hugepages-1Gi: 2Gi
volumeMounts:
- name: hugepage
mountPath: /hugepages
- name: varrun
mountPath: /var/run/dpdk
readOnly: false
- name: iss-path
mountPath: /opt/intel/system_studio_2019/
- securityContext:
privileged: false
capabilities:
add:
- SYS_ADMIN
- IPC_LOCK
- SYS_NICE
command:
- sleep
- infinity
tty: true
stdin: true
image: oran-release-du:4.0
name: oran-fapi
resources:
requests:
hugepages-1Gi: 4Gi
memory: "4Gi"
limits:
hugepages-1Gi: 4Gi
memory: "4Gi"
volumeMounts:
- name: hugepage
mountPath: /hugepages
- name: iss-path
mountPath: /opt/intel/system_studio_2019/
- name: varrun
mountPath: /var/run/dpdk
readOnly: false
- securityContext:
privileged: false
capabilities:
add:
- SYS_ADMIN
- IPC_LOCK
- SYS_NICE
command:
- sleep
- infinity
tty: true
stdin: true
image: oran-release-du:4.0
name: testmac
resources:
requests:
memory: "4Gi"
limits:
memory: "4Gi"
volumeMounts:
- name: hugepage
mountPath: /hugepages
- name: iss-path
mountPath: /opt/intel/system_studio_2019/
- name: varrun
mountPath: /var/run/dpdk
readOnly: false
volumes:
- name: hugepage
emptyDir:
medium: HugePages
- name: varrun
emptyDir: {}
- name: iss-path
hostPath:
path: /opt/intel/system_studio_2019/
Create the pod using kubectl
kubectl create –f oran-release-du.yaml
After the pod/containers are running, login different containers (l1app, fapi, testmac) and start the applications. Open first terminal and execute l1app container, and run l1app
kubectl exec oran-release-du-pod -c l1app -it bash
source /opt/oran_release/phy/setupenv.sh
cd /opt/oran_release/FlexRAN/l1/bin/nr5g/gnb/l1/
./l1.sh -e
After l1app is started, open second terminal, execute oran-fapi container, and run fapi
kubectl exec oran-release-du-pod -c oran-fapi -it bash
source /opt/oran_release/phy/setupenv.sh
cd /opt/oran_release/phy/fapi_5g/bin
./oran_5g_fapi.sh --cfg oran_5g_fapi.cfg
Open third terminal, execute testmac container, and run testmac
kubectl exec oran-release-du-pod -c testmac -it bash
source /opt/oran_release/phy/setupenv.sh
cd /opt/oran_release/FlexRAN/l1/bin/nr5g/gnb/testmac
./l2.sh --testfile oran_bronze_rel_fec_sw.cfg
After testmac container is started and running, you can see the test is running. You can check the terminal of each container to verify the test result.
It was described in the O-RAN SC O-DU Low project document, get through below link https://docs.o-ran-sc.org/projects/o-ran-sc-o-du-phy/en/latest/index.html