TOSCA Components

You can modify the node components and CaaS components in TOSCA for different Kubernetes VIMs.

To support various network functions, the Worker nodes may require a customization in the TOSCA. These customizations include the kernel-related changes, custom packages installations, network adapter, SRIOV, DPDK configurations, and CPU Pinning of the Worker nodes on which you deploy the network functions.

Node Components

Kernel: The Kernel definition uses multiple arguments that require a customization.

kernel_type: Kernel type for the worker nodes. The kernel types are:
- Linux RealTime (linux-rt)
- Linux Non-RealTime (linux)
The kernel type depends on the network function workload requirement. The required Linux version is downloaded from TDNF repo[VMware Photon Linux] during customization.
kernel type
```
infra_requirements:
  node_components:
    kernel:
      kernel_type:
        name: linux-rt
        version: 4.19.132-1.ph3
```

kernel_args: Kernel boot parameters for tuning values that you can adjust when the system is running. These parameters configure the behavior of the kernel such as isolating CPUs. These parameters are free form strings. They are defined as 'key' → name of the parameter and optionally 'value' → if any arguments are provided.

kernel_args

infra_requirements:
  node_components:
    kernel:
      kernel_args:
        - key: nosoftlockup
        - key: noswap
        - key: softlockup_panic
          value: 0
        - key: pcie_aspm.policy
          value: performance
        - key: intel_idle.max_cstate
          value: 1
        - key: mce
          value: ignore_ce
        - key: fsck.mode
          value: force

Huge Pages

infra_requirements:
  node_components:
    kernel:
      kernel_args:
        - key: default_hugepagesz
          value: 1G
        - key: hugepagesz
          value: 1G
        - key: hugepages
          value: 17
Note:
i.   This order should be maintained.
ii.  Nodes will be restarted to set these values
iii. supported hugepagesz are 2M | 1G

isolcpus

infra_requirements:
  node_components:
    kernel:
      kernel_args:
        - key: isolcpus
          value: 2-{{tca.node.vmNumCPUs}}
Note: TCA will replace the {{tca.node.vmNumCPUs}} with vCPUs configured on the worker node.

kernel_modules: To install any kernel modules on Worker nodes. For example, dpdk, sctp, and vrf.
Note: When configuring dpdk, ensure that the corresponding pciutils package is specified under custom_packages.

dpdk
```
infra_requirements:
  node_components:
    kernel:
      kernel_modules:
        - name: dpdk
          version: 19.11.1
```
For a details on supported DPDK versions, see Supported DPDK and Kernel Versions.

custom_packages: Custom packages include the lxcfs, tuned, pci-utils, ptp. The required packages are downloaded from TDNF repo[VMware Photon Linux] during customization.

custom_packages

infra_requirements:
  node_components:
    custom_packages:
       - name: pciutils
         version: 3.6.2-1.ph3
       - name: tuned
         version: 2.13.0-3.ph3
       - name: linuxptp
         version: 2.0-1.ph3
Note: Make sure these packages are available on VMWARE TDNF Repository

additional_config: Helps in the additional customization on node. For example, tuned.

Note: While configuring tuned, ensure that the corresponding tuned package is specified under custom_packages

tuned

infra_requirements:
  node_components:
    additional_config:
      - name: tuned   # <--- for setting tuned
        value: '[{"name":"custom-profile"}]'    # <--- list of profile names to activate.

file_injection: Inject the configuration files inside the nodes.

file_injection

infra_requirements:
  node_components:         
    file_injection: 
      - source: file
        content: ../Artifacts/scripts/custom-tuned-profile.txt  #<-- File path location which is embedded in CSAR
        path: /etc/tuned/custom-profile/tuned.conf  #<-- Target location of the configuration file. Location should align with name of the profile.
      - source: file 
        content: ../Artifacts/scripts/cpu-partitioning-variables.txt  #<-- File path location which is embedded in CSAR
        path: /etc/tuned/cpu-partitioning-variables.conf #<-- Supporting files for the main configuration file.

isNumaConfigNeeded: This feature tries to find a host and a NUMA node that can fit the VM with the given requirements and assign it. It is useful for high-performance profile Network Functions such as DU, which require a high throughput. This sets CPU and Memory reservations to maximum on the Worker node. It sets the affinity for the Worker node cpus to the ESXi cpus.
isNumaConfigNeeded
```
infra_requirements:
   node_components:
     isNumaConfigNeeded:  [true | false]
```
latency_sensitivity: For Network Functions that require a high-performance profile with low-latency such as DU, CU-CP, CU-UP, and UPF. These functions require the node latency sensitivity set on vSphere.
Note: Node restarts after customization.

latency_sensitivity
```
infra_requirements:
  node_components:         
    latency_sensitivity:  [high | low]
```
passthrough_devices: For adding PCI devices. For example, ptp.
Note: While specifying passthrough device configurations, ensure that the corresponding linuxptp package is specified under custom_packages.

passthrough_devices
```
 infra_requirements:
   node_components:
     passthrough_devices:
       - device_type: NIC
         pf_group: ptp
Note: For now the values are hardcoded
```

network: Creates network adapters on the nodes. For SRIOV, the given resource name will be allocatable resource on the node.

Network

 infra_requirements:
   node_components:
     network:
        devices:
          - deviceType:    # <-- Network Adapter type [sriov]
            networkName:   # <-- Input label for User Input to provide Network while NF Instantiation. Refer below section how to define these input
            resourceName:  # <-- This is the label the device will be exposed in K8s node.
            dpdkBinding:   # <-- The driver this device should used. If not mentioned, then default OS driver will be used.
            count: 3       # <- Number of adapters required.
Note:
    1. vmxnet3 not supported in TCA
    2. for 'networkName' refer below section
    3. dpdkBinding
       - igb_uio
       - vfio-pci
    4. Make sure to have 'pciutils' custom packages and 'dpdk' kernel modules.

For SRIOV network adapters, when initiating, add the following:

VnfAdditionalConfigurableProperties

tosca.datatypes.nfv.VnfcAdditionalConfigurableProperties.lmn:
  derived_from: tosca.datatypes.nfv.VnfcAdditionalConfigurableProperties
  properties:
    F1U: # <--- label that is provided infra_requirements.node_components.network.devices.networkName
      required: true
      propertyName: F1U # <--- label that is provided infra_requirements.node_components.network.devices.networkName
      description: ''
      default: ''
      type: string
      format: network        #<- to show the network drop-down menu
 
 helm-abc:
    type: tosca.nodes.nfv.Vdu.Compute.Helm.helm-abc
    properties:
      :
      configurable_properties:
        additional_vnfc_configurable_properties:
          type: tosca.datatypes.nfv.VnfcAdditionalConfigurableProperties.lmn
          :
          :
          F1U: '' # <-- Same label provided above

caas_components

You can configure CaaS components, such as CNI, CSI, Helm for the Kubernetes. You can install CNI plugins on Worker nodes during CNF instantiation. Provide CNIs such as SRINOV in Cluster Configuration in the CaaS Infrastructure.

infra_requirements:
  caas_components:
    - name: sriov
      type: cni