Customers (especially Telco and large enterprises) need fast, controllable datapaths for both containers and VMs without tying deployments to SR-IOV hardware. Today, OCP's default path (OVN-Kubernetes on OVS kernel datapath) struggles to meet deterministic low-latency/throughput targets for NFV/CNF and migration scenarios. This Feature tracks an engineering research initiative to build a userspace networking POC and evaluate whether it's viable to productize.

This research PoC is intended to complement the existing OpenShift networking components. It does not replace or change the default networking and any potential outcome would be tightly integrated with current OCP components. 

Business problem

• Provide a high-performance datapath for pods and VMs that keeps policy control (ACLs, services, observability) and reduces dependence on SR-IOV NICs/VFs.
• Enable customer migrations (VMware, OSP with ovs-dpdk) and address Telco VNFs/CNFs that need consistent latency and high PPS with clear resource sizing.

This Feature does not commit to GA; it establishes feasibility, performance, and risk to inform a go/hold/no-go decision.

Objectives

• Build an installable PoC image/profile enabling userspace primary networking on OCP.
• Demonstrate functional parity for core networking (routing, services/LB, SG/ACL, DHCP/RA, IPv4/IPv6).
• Produce performance results for pod↔pod and VM↔VM, cross-node: latency, throughput (Gbps/PPS), and CPU/memory footprint.
• Identify gaps (OVS/DPDK/OVN-K/KubeVirt/VDUSE), required patches and operational caveats.
• Deliver a productization recommendation (with options: Tech Preview, GA later, or do not proceed) plus supportability plan.

Customers / Markets 

• Telco (VNF/CNF) requiring deterministic latency and high PPS with policy control.

• Enterprises using SR-IOV today but needing a policy-rich, more flexible alternative.

• VMware to OCP migrations and OSP(ovs-dpdk) to OCP transitions.

Deliverables

• PoC enablement (installer/profile/manifests) and how-to.

• Benchmark report: latency (P50/P99), throughput (Gbps, PPS), CPU/memory/utilization, plus config used (PMD cores, MTU, queues).

• Gap analysis: list of required upstream changes/workarounds and their status.

• Decision brief: recommendation (TP/GA later/no-go) with risks and supportability.

Acceptance criteria

• PoC deploys and runs with OVS-DPDK userspace primary networking for pods (VDUSE) and VMs (vhost-vDPA) with OVN-Kubernetes, IPv4/IPv6.

• Published performance results (methodology + configs) for pod to pod and VM to VM across nodes:
  Latency (incl. P99), Throughput (Gbps/PPS), CPU/Memory utilization at target load.

• Documented gaps & risks with an engineering recommendation (TP/GA/no-go) and proposed next steps.
• VM Live migration should functionaly work