When choosing a Wi-Fi architecture, the AP model gets debated while the real decision is overlooked: who will manage the network, and where will client traffic flow? A hardware or virtual controller, a cloud-managed platform and a controllerless (distributed) architecture differ significantly in behavior during outages, scale, licensing cost and operating model. A wrong choice shows up not on installation day but at the first WAN outage or the first subscription renewal. Below we compare the three approaches through the control, management and data planes and give a decision tree based on branch structure.
Control, management and data planes in Wi-Fi architectures
Compare architectures not by their names in sales brochures but by where the three planes run:
- Management plane: configuration, firmware distribution, monitoring, reporting and alarms. It can run on a controller, in the cloud or on a “virtual controller” within the AP group.
- Control plane: RF management (channel/power), client association decisions, distribution of roaming keys, authentication forwarding. Functions that need to run in near real time are here.
- Data plane: the path client traffic takes. It’s either bridged locally to a VLAN at the AP or carried to a central point through a tunnel.
The phrase “cloud-managed” often describes only that the management plane is in the cloud; most control functions stay on the APs, and the data plane stays local. In a “controller-based” architecture, the control plane is on the controller, and the data plane can be tunneled or local depending on the design choice. All of the outage and performance analysis below rests on this distinction. Don’t decide without making it.
Controller-based architecture: central tunnel and local bridging
In a controller-based architecture, APs build a tunnel to a hardware or virtual controller. The CAPWAP protocol used by Cisco is defined in RFC 5415; UDP 5246 is used for the control channel and UDP 5247 for the data channel. Other vendors may use their own GRE- or IPsec-based tunnel structures. If APs can’t join the controller, the first place to look on the console is whether these ports are open on the firewall in between, and whether the AP really learns the controller address (via a DHCP option or DNS record).
Central tunnel
Client traffic is carried from the AP to the controller through a tunnel and exits to the VLAN there. Advantages: client VLANs exist only on the core or data center switch the controller is connected to, the access ports the APs are connected to stay simple, the IP address is preserved during roaming even across floors and buildings, and security policy is applied at a single point. Limitations: the controller’s data processing capacity can become a bottleneck; at branches, traffic going over the WAN to the center and back (hairpinning) creates latency and bandwidth cost.
Local bridging
The AP sends client traffic directly out of the switch port it’s connected to with a VLAN tag; the controller handles only control and management functions. On Cisco, this is known as FlexConnect local switching. It suits branches, but requires client VLANs to be carried as a trunk on access switches. A mistake you may encounter comes up here: the SSID’s VLAN isn’t in the AP port’s trunk allowed list, so the client connects but can’t get an IP. For access port configuration and the VLAN plan, see the VLAN design in enterprise networks article.
In this architecture, the real design topic is controller redundancy. What APs will do when a single controller is lost (restart, fail over to a backup controller, continue in local mode) must be designed in advance. An active–standby controller pair or N+1 redundancy are common approaches. A single controller is a single point of failure.
Cloud-managed architecture: advantages and dependencies
In a cloud-managed architecture, APs (and often switches and firewalls) run registered to the vendor’s cloud platform. Configuration, monitoring and firmware management are done from the web interface and via API. The data plane is usually local.
Pros:
- There’s no controller hardware, redundancy or software upgrade burden at the center.
- Commissioning a new branch is fast: the AP is added to the platform by serial number, a template configuration is applied and the device is shipped to the site.
- All locations in the enterprise network infrastructure are monitored from a single screen; client history and event logs are kept centrally.
- Inventory and configuration automation become easier with APIs.
Dependencies:
- Management access depends on the internet and on the vendor’s service continuity.
- Some features (cloud RADIUS, cloud captive portal, identity-based PSK lists) may require live cloud access to work.
- Event and client data are kept in the vendor’s data center; data residency and personal data processing terms should be evaluated with respect to KVKK.
- Configuration options are limited to what the platform offers; detailed CLI settings aren’t always available.
The last item looks small. It isn’t. Not being able to get into the AP and run debug commands during a fault directly lengthens troubleshooting time; before choosing the platform, try out which diagnostic tools it offers.
What happens during an internet outage?
This is the most misunderstood topic in choosing an architecture. The table below shows the general tendency; the exact behavior varies by vendor, software version and SSID configuration, and should be tried in the acceptance test.
| Function | Controller (on-premises) | Cloud-managed | Controllerless |
|---|---|---|---|
| Existing clients | Unaffected | Usually continue with local bridging | Unaffected |
| 802.1X with local RADIUS | Works | Usually works | Works |
| Cloud RADIUS / cloud portal | Not applicable | May be affected | Not applicable |
| Configuration changes | Possible | Not possible | Possible |
| Monitoring and event logging | Continues | May be incomplete during the outage | Continues |
In a controller-based architecture, the critical dependency isn’t the internet but the WAN link between the AP and the controller. A branch running in central tunnel mode can lose wireless service entirely when the WAN drops (if there’s no fallback-to-local-mode feature). The internet is up, Wi-Fi is down. For branch WAN design and a backup line, the dual internet line and failover article provides complementary information.
The effect of the licensing and subscription model on the life cycle
Comparing cost by one-time hardware price is misleading. The right comparison is total cost over the AP’s planned service life (5–7 years in most organizations):
- Controller-based: controller hardware or virtual controller resources, a backup controller, AP licenses, a support contract and labor for controller software upgrades.
- Cloud-managed: an annual or multi-year subscription per AP. On most platforms the subscription covers software updates and support; the real question is what happens when the term expires. On some platforms only management access is restricted; on others, devices may stop passing traffic when the grace period ends.
- Controllerless: usually the lowest initial cost; as scale grows, an additional management platform may be needed for central visibility.
Clarify in writing before buying: device behavior at subscription expiry, portability of licenses between devices, how end-of-sale and end-of-support dates are announced, and whether APs can be converted to another management mode (e.g. a local controller) if the platform is changed. The last point directly determines vendor lock-in. A verbal answer isn’t enough.
Comparison table
| Criterion | Controller-based | Cloud-managed | Controllerless |
|---|---|---|---|
| Scale | Large campus | Many branches | Single site, small–medium |
| Data plane | Tunnel or local | Mostly local | Local |
| Roaming within the campus | Strong, IP preserved | Depends on the local VLAN design | Enough at small scale |
| Operating effort | High | Low | Low to medium |
| Cost structure | Hardware + license + support | Periodic subscription | Mainly hardware |
| Internet dependency | None | Exists for management | None |
A decision tree by scale and branch structure
- How many locations are there? For a single location and a scale of up to a few dozen APs, controllerless or cloud-managed solutions are considered. If there are many branches, go to step 2.
- Is there on-site IT staff at the branches? If not, zero-touch commissioning and central monitoring are the priority; a cloud-managed architecture is a strong candidate.
- Is there a large campus or a heavy roaming need (hospital, production, warehouse)? If so, a hybrid structure with a controller-based central tunnel on the campus and local bridging at the branches is considered.
- Is it acceptable for management data to be held abroad, or for there to be an internet dependency? If not, an on-premises controller or an on-premises management platform is preferred.
- Does the team work with CLI and automation tools, or with a web dashboard? The operating model and existing skills determine the sustainability of the choice. An architecture nobody can maintain isn’t a good architecture.
When setting branch standards, the wireless architecture should be consistent with switch, WAN and firewall choices; the standard branch network design article guides this consistency. Evaluating the architecture choice together with existing infrastructure and the operating model falls within enterprise network consulting, and installing the chosen architecture and its RF tuning within wireless network installation and optimization.
Frequently asked questions
If the internet goes down on cloud-managed Wi-Fi, do users lose their connection?
On SSIDs that use local bridging and whose authentication relies on local resources, APs usually keep running with the last configuration they received. SSIDs that use cloud-based RADIUS, a cloud captive portal or a PSK list held in the cloud may be affected. Because the behavior varies by vendor and feature, try the outage scenario in the acceptance test.
In a controller-based architecture, does all traffic pass through the controller?
Not necessarily. In central tunnel mode, client traffic is carried to the controller; in local bridging mode (such as FlexConnect local switching on Cisco), traffic exits directly at the switch the AP is connected to, and the controller handles only control and management functions. Different modes can be used per SSID on the same controller.
What happens if the cloud subscription isn’t renewed?
It varies by vendor: on some platforms only management and reporting access is lost; on others, devices may stop passing traffic after a grace period ends. Clarify this behavior in writing in the contract and in the vendor’s licensing documentation before buying.
Is controllerless (distributed) Wi-Fi enough for small companies?
For single-site networks of up to a few dozen APs without advanced policy needs, it’s usually enough. In multi-branch structures, because the need for central visibility, configuration templates and life cycle management grows, controller or cloud management becomes more suitable.