Configuring a NFV deployment takes a bit of coordination and its difficult to do manually.
As an Telco company I would like to have an easy way to deploy a VNF worker configured with the following:

• ovs dpdk fast data-path on required interfaces
• all worker host required drivers installed in CoreOS to access the fast data-path nics (PMD-drivers).
• all required OpenStack resources defined (flavors, security groups, etc..)

The system should consist of nfv workers which meet the requirements of the worker nodes in the below diagram:
https://coreos.slack.com/files/U012XCAEUKX/F028FUM6R5M/osasinfra_telco_openshift_perf_lab_.jpeg

Requirements:

• OpenStack is configured to use OVS-DPDK for its internal OVS switches, as described in the official documentation [1]. This includes the allocation of memory hugepages, NUMA allocation , and CPU pinning on the OpenStack Compute node required for the deployment. All described in the documentation listed under [1]
  The OpenStack deployment should expose vms and flavors with the following characteristics:
• TBD - list all the requirements for the vm to be used as an OpenShift node. [2] (for example, driver installation, hugepages requirements, and other necessary software that needs to be installed and configured on the worker node so it can access the nics connected to the fast data-path.
• The OpenShift cluster will NOT reach into the OpenStack deployment to make any configuration which require administrative privileges. All OpenStack configurations requiring administrative privileges will be done by the user.

System characteristics:

• Control-plane attaching workers to controllers will NOT be accelerated by OVS-DPDK. i.e. the OVN switches inside the CoreOS based worker which attach the pods to the control-plane will NOT implement OVS-DPDK.
• Workers will have at least two interfaces attached to them that are exposed to the pods using a PMD (Poll Mode Driver) and are in turned accelerated by the OVS-DPDK switch in the OpenStack Compute node. In total the worker should have three interfaces. One attached to the cluster control-plane which will NOT be accelerated using and is exposed to the node as a virtio or host_user, two interfaces attached to low latency (fast data-path) networks, which will be exposed to the vm as host_user interfaces.

limitations:

• The network path of the control plane, which connects the workers to the controllers on the machineNetwork (for openshiftSDN), cannot be accelerated. Thus traffic flowing on the OpenShift control-plane will not be accelerated by ovs-dpdk.

Acceptance Criteria:

• Document describing how to deploy the system which may be used later by the customer as a guide for developing an operator to automate the process
• Following the document, a system like the one shown in Fig 2, is deployed and "testpmd" pods can be deployed on to the workers, and performance test is executed. "testpmd" pods are pods that can run testpmd using the fast data-path [3].
• Aspirational, the NICs used to attache the pods to the fast data-path are VF from an SR-IOV device which was configured by OpenStack administrator to expose them to the worker nodes.

References:
[1] https://access.redhat.com/documentation/en-us/red_hat_openstack_platform/13/html/network_functions_virtualization_planning_and_configuration_guide/part-dpdk-configure
[2] https://docs.openstack.org/neutron/queens/admin/config-ovs-dpdk.html
[3] https://docs.openvswitch.org/en/latest/howto/dpdk/