SDN/NFV Compute Node: Virtualised Network Functions, DPDK & Telco Cloud
Published June 21, 2026 • 7 min read • Telecom Insights
The SDN/NFV Compute Node is the physical server that hosts virtualised network functions (VNFs) and cloud-native network functions (CNFs) — replacing dedicated telecom appliances with software running on commercial off-the-shelf (COTS) hardware. This transformation — driven by ETSI NFV standards since 2012 — has reshaped the economics of telecom networks, enabling operators to deploy, scale, and upgrade network functions with cloud-era agility.
Key Takeaway: A single NFV compute node can host a 5G CU-UP, UPF, SMF, and AMF simultaneously — functions that previously required racks of dedicated hardware. The orchestrator can spin up new instances in seconds to handle traffic spikes, then tear them down when demand subsides.
NFV Infrastructure (NFVI) Architecture
The ETSI NFV reference architecture defines: NFVI (NFV Infrastructure) — the physical compute, storage, and network resources plus the virtualisation layer (hypervisor/container runtime); VNFs (Virtualised Network Functions) — the software implementations of network functions running as VMs (KVM/QEMU) or containers (Docker, containerd); NFV MANO (Management and Orchestration) — the framework for VNF lifecycle management; and VIM (Virtualised Infrastructure Manager) — the cloud platform (OpenStack, Kubernetes) managing the NFVI resources. The compute node runs the VIM and hosts the VNFs/CNFs.
Data Plane Acceleration: DPDK, SR-IOV, and SmartNICs
Virtualised network functions face a fundamental challenge: standard Linux kernel networking (interrupt-driven, sk_buff-based) cannot handle the 10–100 Gbps per-VNF throughput required. Three acceleration technologies bridge the gap. DPDK (Data Plane Development Kit): a set of userspace libraries and poll-mode drivers that bypass the kernel, processing packets directly from the NIC via DMA with run-to-completion or pipeline models — achieving 100+ Mpps on a single CPU core. SR-IOV (Single Root I/O Virtualisation): exposes the NIC's physical functions as multiple virtual functions, each directly assigned to a VNF via PCIe passthrough — near-native performance with hardware isolation. SmartNICs and DPUs: offload the entire data plane to a programmable network processor on the NIC — the CPU runs control-plane code while the SmartNIC processes packets at 200 Gbps line rate with deterministic latency.
VNF to CNF Evolution
The industry is transitioning from VNFs (monolithic VMs with embedded OS) to CNFs (Cloud-native Network Functions) — containerised microservices orchestrated by Kubernetes. CNFs decompose a network function into independently scalable services: the 5G SMF might become separate containers for PFCP handling, session state management, and charging interface. Benefits include: faster instantiation (containers start in <1 second vs minutes for VMs), higher density (100+ containers per node vs 10–20 VMs), CI/CD deployment (GitOps-driven continuous delivery), and auto-scaling via Kubernetes HPA. The Service Mesh (Istio/Linkerd) provides the communication, observability, and security fabric for inter-CNF communication.
MANO: ONAP and Orchestration
ONAP (Open Network Automation Platform) — the Linux Foundation's flagship NFV orchestration project — implements the ETSI MANO framework. ONAP's components: SDC (Service Design and Creation) — design VNF/CNF descriptors (TOSCA/YAML); SO (Service Orchestrator) — end-to-end service lifecycle; VF-C (VF Controller) — VNF lifecycle (instantiate, scale, heal, terminate); DCAE (Data Collection, Analytics, and Events) — closed-loop automation (collect telemetry, detect anomalies, trigger policy-driven actions); and Policy Framework — declarative policies for scaling, placement, and healing. ONAP integrates with multiple VIMs (OpenStack, Kubernetes, VMware) and SDN controllers (ONOS, ODL).
Performance Benchmarking and Assurance
Telco workloads demand deterministic performance that enterprise cloud platforms struggle to deliver. Key NFVI performance metrics: throughput (line rate at all packet sizes including IMIX), latency (P99 <500 µs for UPF N3/N6 processing), jitter (<50 µs), packet loss (zero loss under rated load), and noisy neighbour resistance (guaranteed performance regardless of co-located workloads). Technologies enabling telco-grade performance: CPU pinning and isolation (dedicated cores for VNFs, isolated from host OS), huge pages (1 GB pages for DPDK memory), NUMA awareness (VNF pinned to local NUMA node's cores and NIC), real-time kernel (PREEMPT_RT patch for bounded latency), and CMT (Cache Monitoring Technology) and MBM (Memory Bandwidth Monitoring) for resource contention detection. The OPNFV (Open Platform for NFV) project provides reference testing frameworks.
The SDN/NFV compute node is the foundational building block of the telco cloud — transforming the operator's network from a collection of special-purpose hardware boxes into a programmable, software-defined platform that can innovate at software speed.
