skip to content

In enterprise Wi-Fi, how do autonomous, controller-based and cloud-managed access points differ in where configuration, control and user traffic are handled?

level: juniorimportance: must knowfreq 58%

answer

  1. three planes, three homes
  2. each AP configured on its own
  3. WTP and AC over CAPWAP
  4. management plane in a hosted service
  5. what survives losing the manager

basics

~20 s

An autonomous AP keeps configuration, control and forwarding to itself. A controller-based AP takes its configuration and control decisions from a controller over CAPWAP. A cloud-managed AP is configured from a hosted service but normally controls and forwards locally.

solid answer

~50 s

Split an AP's work into three planes: **management** (configuration, firmware, monitoring), **control** (admitting clients, keys, radio decisions) and **data** (forwarding user frames). An **autonomous** AP does all three itself, so every AP is configured separately and none knows what its neighbours are doing. A **controller-based** AP, a WTP in CAPWAP terms, joins an access controller (AC), takes its configuration and firmware from it, and leaves client admission, key management and radio management to it; user traffic is either tunnelled to the controller or bridged locally at the AP. A **cloud-managed** AP keeps control and forwarding on the AP and moves the management plane to a hosted service reached over the internet, so losing that link usually freezes changes and monitoring while the SSIDs keep running. The choice turns on scale, consistency and what must survive a lost link.

go deeper

for a junior

Name the three models and say, for each, where an AP gets its configuration and where a client's traffic goes.

for a middle

Explain the management, control and data planes and which of them each model centralises, using CAPWAP's WTP and AC terms for the controller case.

for a senior

Show what each model loses when its manager is unreachable, and which client steps depend on the controller or the hosted service at connect time.

for a principal

Frame the choice as scale, consistency, failure domains and operating cost, and say what evidence you would want before standardising one model across many sites.

## Three planes, three homes Every access point (AP) does three kinds of work, and the three management models differ only in **where each kind of work lives**: - **Management plane**: the configuration (SSIDs, security settings, VLANs, radio settings), firmware upgrades, monitoring and logs. - **Control plane**: the decisions taken while the network runs, such as whether to accept an association, who runs 802.1X and key management, and which channel and transmit power each radio uses. - **Data plane**: moving user frames between the radio and the wired network, either by bridging them onto a local VLAN or by tunnelling them somewhere else first. ## Autonomous APs An **autonomous** (standalone) AP holds all three planes itself. A `hostapd.conf` file for hostapd, the open-source reference implementation, is a good picture of one: SSID, security, channel, WMM and roaming options all sit in one local file that only that AP reads. - It works well for a handful of APs and needs nothing else on the network. - Every AP is configured on its own, so ten APs mean ten configurations kept in step by hand or by your own tooling, and **configuration drift** is the normal failure. - Radio decisions are local. hostapd's automatic channel selection (`channel=acs_survey`) surveys the air when the radio starts and picks the quietest channel it can see, with no knowledge of what its neighbours chose a minute later. - An autonomous AP can still offer WPA2-Enterprise or WPA3-Enterprise: it simply acts as the RADIUS client itself. ## Controller-based APs and CAPWAP In a **controller-based** design the APs are managed by an **access controller**. CAPWAP (RFC 5415, with the IEEE 802.11 binding in RFC 5416) calls the AP a **Wireless Termination Point (WTP)** and the controller an **Access Controller (AC)**. RFC 5416 states the goal plainly: move access control, user authentication and authorization, mobility management and radio management from the single AP to a centralized controller. 1. The AP finds a controller (static address, DHCP, a DNS SRV record, or broadcast or multicast discovery). 2. It builds a DTLS-protected control session to the controller's UDP port 5246. 3. It joins, downloads firmware if its version differs, and receives its configuration. 4. From then on the controller pushes WLANs, channel and power settings and station state to it. In CAPWAP's 802.11 binding the AP always generates beacons and probe responses, while 802.1X/EAP and key management run on the controller, which therefore acts as the RADIUS client. User traffic is either **tunnelled** to the controller over CAPWAP's data channel or **bridged locally** at the AP. ## Cloud-managed APs A **cloud-managed** AP keeps the control and data planes on the AP and moves most of the **management plane** to a hosted service that the AP reaches over an outbound internet connection. - Configuration, firmware and monitoring come from the service; the AP keeps running the last configuration it received. - The management protocol is usually the provider's own rather than a standard such as CAPWAP, so APs and service tend to be bought together. - Coordinated functions that need a wide view, such as channel and power optimisation across a site, are computed by the service and pushed down. - Losing internet access normally stops changes and monitoring, not forwarding. Anything the AP must ask the service at connect time, such as a hosted splash page, is the exception to check. ## Side by side | | Autonomous | Controller-based | Cloud-managed | |---|---|---|---| | Configuration lives | On each AP | On the controller | In the hosted service, cached on the AP | | Admission and keys | On the AP | On the controller (CAPWAP binding) | On the AP | | Radio decisions | Each AP alone | Controller, across APs | Often computed by the service, applied by the AP | | User traffic | Bridged at the AP | Tunnelled to the controller or bridged at the AP | Bridged at the AP | | Manager unreachable | Nothing to lose | Depends on what the controller owned | Changes and monitoring stop | ## Choosing in practice 1. **Count APs and sites.** A few APs in one office can stay autonomous; dozens or hundreds need a central manager. 2. **List what must survive a lost link.** Whatever the manager performs at connect time fails with it. 3. **Decide where traffic must be inspected.** Central inspection argues for tunnelling some SSIDs; local services argue for local bridging. 4. **Match the operations model.** Running controllers is work your team owns; a hosted service is a recurring subscription and a dependency on a provider. The models are not ranked from worst to best. They move the same three planes to different homes, and the right home depends on scale and on which failure the estate can tolerate.

  • Can an autonomous AP offer WPA2-Enterprise without a controller?
    Yes. An autonomous AP can act as the RADIUS client itself, relaying 802.1X/EAP to the authentication server and running key management locally; hostapd does exactly that. What it lacks is coordination with other APs, not the ability to authenticate.
  • Is a controller-based AP useless when it cannot reach its controller?
    CAPWAP sends a WTP that loses its AC back to discovering one; whether it keeps serving clients meanwhile is the implementation's behaviour, not the protocol's. Locally bridged traffic for clients already connected may carry on, but any step the controller owns, such as 802.1X and key management in CAPWAP's 802.11 binding, fails for new clients unless the AP has a local fallback.
  • Does cloud management mean user traffic travels through the provider's service?
    Normally not. The hosted service carries configuration, monitoring data and firmware; the AP bridges user frames onto the local network. Traffic leaves for the provider only if you enable a feature that needs it, such as a hosted splash page, so check each feature's dependency before assuming an internet outage is harmless.

A chain of shops: autonomous APs are shops each run by its own manager with its own rulebook; controller-based APs are shops that phone head office to approve each new customer; cloud-managed APs get the rulebook posted from head office and keep trading by it when the phone line is down, they just cannot receive a new edition.

  • Cloud-managed means all user traffic is backhauled through the provider's service.
  • Autonomous APs cannot offer 802.1X because only a controller can be a RADIUS client.
  • A controller only pushes configuration and plays no part once clients connect.
  • Losing the internet link immediately takes every cloud-managed SSID off the air.
  • Autonomous APs agree a channel plan with their neighbours automatically.