Why 802.11 Needs a Forwarding Plane, and Where DPDK Fits

Robert McMahon (founder and CTO, Umber Networks) makes the case that 802.11 has the same structural problem Ethernet had before switching: no deterministic forwarding plane. The MAC layer is stochastic by design, firmware-owned, and invisible to every scheduling decision above it. His argument is that DPDK should become the forwarding plane for the wireless edge, the same way it runs the wired edge today. The core technical problem: the transmit opportunity (TX op) is the unit of work in 802.11, not the Ethernet frame. With listen-before-talk and stochastic backoff, the medium access cost alone limits throughput to 10,000 transmits per second regardless of PHY rate. Aggregation (A-MPDU) recovers capacity but loses 64 packets on a single retransmit. Five active transmitters gives a 50% probability of collision. None of this is visible to IP, TCP, or any Linux traffic control mechanism, it happens underneath, in vendor firmware, with no shared state across access points. McMahon's proposed architecture splits the PHY from the MAC, moves scheduling to a host-based DPDK forwarding plane connected to radio heads over PCIe, and gives the host building-wide RF state. The firmware change required per chip is roughly three months of engineering work. The barrier is business model alignment, not technical complexity. Timestamps 0:00 Introduction: Bob McMahon, Umber Networks, background in Wi-Fi chip testing at Broadcom 1:16 Why 802.11 and DPDK matter for wireless networks going forward 2:36 802.11 is not Ethernet: the abstraction that has run its course 4:33 TX op as the unit of work: 10,000 transmits/sec ceiling, stochastic backoff, A-MPDU aggregation 9:21 Five active transmitters, 50% collision probability: the MAC is the scarce resource 10:52 Multi-destination aggregates and the scheduling problem 13:41 The PHY layer: why the FI is the Ferrari and the MAC is the garbage truck 16:33 Multiple APs, autonomous clocks, noise floor interference, no shared RF state 19:00 The verdict: Wi-Fi cannot be scheduled, determinism is impossible with current architecture 20:35 Historical forwarding plane evolution: IP routing, Ethernet switching, and what comes next 22:57 802.11 as the third forwarding plane problem 24:20 Proposed architecture: split PHY from MAC, host-based scheduling, DPDK as forwarding plane 26:31 Required firmware parameters: EDCA, MCS edges, backoff, channel state information 28:06 Centralised RF state: energy detect, NAV, host PMD pulling from local DRAM 30:00 MCS index graph: testing edges between nodes, not just the table 33:30 Summary: Wi-Fi needs a forwarding plane the way Ethernet needed switching 34:24 Q&A: getting firmware from existing vendors vs. building new chips 38:09 Q&A: why autonomous APs cannot solve the coordination problem 39:53 Q&A: lessons from 5G, O-RAN splits, raw IQ samples over 100G links 44:21 Q&A: interference as a MAC-level design choice, not a physics constraint 45:49 Q&A: complexity compared to cellular base stations 48:35 Q&A: firmware engineering effort to implement the host-driven model (~3 months per chip) 49:28 Q&A: existing DPDK device API vs. new API requirements 52:03 Q&A: latency as the real user experience metric, L4S, 1ms target within a building 54:10 Q&A: kernel vs. user space for the forwarding plane DPDK (Data Plane Development Kit) is an open source set of libraries and drivers for fast packet processing. It runs on x86, Arm, and PowerPC across CPUs, GPUs, and DPUs from Intel, NVIDIA, AMD, Arm, Marvell, NXP, and others. DPDK is used in production for 5G core networks, cloud infrastructure, scientific computing (including CERN), financial trading systems, and AI workloads. The project is hosted by the Linux Foundation. DPDK member companies: Intel, Red Hat, NVIDIA, AMD, Arm, Marvell, NXP, Ericsson, Microsoft, Huawei, ZTE Get involved https://www.dpdk.org/contribute/ https://www.dpdk.org/review-your-firs... DPDK Dispatch newsletter (quarterly updates, developer spotlights, user stories)   / dpdk-dispatch-7340726346899197952   Learn more about DPDK https://www.dpdk.org DPDK Summit Stockholm 2026 playlist    • DPDK Summit Stockholm 2026 | Talks, Interv...   #DPDK #DPDKSummit #OpenSource