VPP (Vector Packet Processing): High-Performance Networking Framework

VPP (Vector Packet Processing) is an open-source, high-performance, userspace networking framework developed by Cisco (now FD.io / Linux Foundation Collaborative Project). It processes packets in vectors (batches) rather than individually, maximizing CPU cache efficiency and instruction-level parallelism.

Core Architecture

• Vector Processing: Packets collected into vectors (typically 256) → processed through a directed graph of nodes (each node = a function: input, classify, lookup, rewrite, output). Amortizes per-packet overhead (function calls, cache misses, branch prediction).
• Userspace Operation: Runs in user mode (DPDK-based) – bypasses kernel network stack, no context switches, direct NIC access via poll-mode drivers (PMDs).
• Plugin Architecture: Extensible via shared libraries: routing, ACL, NAT, IPsec, load balancing, segment routing (SRv6), telemetry, memif, virtio.
• Control Plane Separation: Data plane (VPP) + control plane (separate process: GoBGP, FRR, custom gRPC/REST API).
• Multi-threading: Worker threads pinned to CPU cores; lock-free ring buffers for inter-thread communication.

Performance Characteristics

• 10–50+ Mpps/core (depending on feature set, packet size, CPU)
• Sub-microsecond latency (data plane)
• Linear scaling with cores (NUMA-aware)

Use Cases

• vCPE / uCPE (universal CPE), vRouter, vBNG, vFirewall
• 5G UPF (User Plane Function), MEC (Multi-access Edge Computing)
• Service Provider: BNG, CGNAT, DPI, load balancing
• Enterprise: SD-WAN edge, microsegmentation
• Cloud/NFV: VPP as VNF/CNF in OpenStack, Kubernetes (FD.io CSIT, VPP K8s Operator)
• Test & Measurement: TRex, VPP-based traffic generators

Getting Started

• GitHub: fdio/vpp (Apache 2.0 license)
• Docs: docs.fd.io
• Binaries: Ubuntu, CentOS, RHEL, Debian packages; Docker images; Vagrant boxes.
• Requires: DPDK-compatible NIC (Intel XL710, X710, E810, Mellanox ConnectX, virtio).

VPP is the data plane engine behind Cisco Meraki, Cisco SD-WAN, Comcast, NTT, and numerous telco/cloud deployments.