Generic RAN DU Workload

A scalable, fully customizable RAN DU workload deployed with kube-burner

Default Configuration

Pod	Number of Pods	Specs	Stress
Guaranteed	1 pod, 2 containers	- 32 CPU, 1 GiB Memory, 16 GiB HP - 2 configmaps and 4 secrets 1 svc	32 threads of 100% CPU stress, 512M virtual memory stress
BestEffort - web_server	4 pods, 2 containers each	- 100 mc CPU, 128 Mib Memory - 2 configmaps and 4 secrets	Exposes 8080 port for probes
BestEffort - curl_app	4 pods, 2 containers each	- 100 mc CPU, 128 Mib Memory - 2 configmaps and 4 secrets - Liveness Probes (every 10 secs)	Kubelet stress with probes, ~250 KB per sec n/w traffic on Primary CNI
BestEffort - kubectl_pods	6 pods, 2 containers each	- 100 mc CPU , 128 Mib Memory - 2 configmaps and 4 secrets - 2 kubectl gets (every 5 sec)	Kube-api-server stress with kubectl get, ~10% increase due to workload
BestEffort - storageio	1 pod, 2 containers	- 100 mc CPU , 128 Mib Memory - 2 configmaps and 4 secrets - 1 hdd stress	Storage-io stress with stress-ng

• Total pods - 15 pods / 30 containers
• 2 config maps and 4 secrets in each pod
• Exec Probes less than 10 in total and frequency >10 secs
• No Exec probes on Gu pod
• Traffic on primary CNI due to workload expected around 350 KB per sec
• kube api-server incremental load due to workload expected to increase by 5-8%

Minimum pre-requisites

• Single Node Openshift Cluster with Performance Profile
• kube-burner v1.16.0+

Steps to run workload on a RAN DU profile Single Node Openshift Cluster

• Clone this repository and run the following.

  export STORAGE_CLASS="" # Defaults to general
  export NETWORKS="" # SRIOV Networks to use for DPDK container if blank then DPDK container will not be added
  export ACCELERATOR_MODEL="" # Used by DPDK deployment if left blank it wont be added
  export REGISTRY="" # For a disconnected cluster, path to mirrored images in internal hub registry
  export NODE_NAME=$(oc get nodes -ojson | jq .items[0].metadata.name -r) # For SNO with performance profile
  export ENABLE_PTP="" # Set to true if node is ptp-enabled 
  export KUBECONFIG="..."
  kube-burner init --config du-intensive.yaml

Steps to run workload with cpu_utilization tests (internal jenkins pipelines)

• Run the MIRROR_SPOKE_OPERATOR_IMAGES stage in ocp-far-edge-vran-deployment pipeline to mirror necessary test images
• Run the cpu_util test using ocp-far-edge-vran-tests pipeline

Additional Workload Features

Enabling PTP Consumer

A PTP consumer can be run on systems with PTP available by setting an env var ENABLE_PTP and requires NODE_NAME.

Running DPDK deployment

Running the DPDK pod requires the defintion of the a NETWORKS env var. $NETWORKS should be a list of objects of the following form:

[
    {
        "network": [1],
        "resource": [2]
    }
]

• [1] The structure for the k8s.v1.cni.cncf.io/networks annotation. It should represent the networks you wish to connect to to the DPDK pod e.g. {"name": "myNetwork", "namespace": "otherNs"}. It will be combined into a json list.
• [2] The key for the resource e.g. openshift.io/pci_sriov_net_myNetwork.