Guest Wi-Fi is usually set up with one sentence: “Let’s open a separate SSID.” A few weeks later, guest phones can see accounting’s printer, the main line fills up with someone’s video stream, and during an incident investigation there’s not a single line of record about which IP was given to which device. A separate SSID doesn’t mean a separate network. Below we explain the architecture of guest access that’s truly isolated from the enterprise network, in implementation order: from SSID and VLAN separation to internet egress, from captive portal options to DHCP scope and logging requirements.
Guest network design goals
A properly built guest network meets four goals at the same time:
- Isolation: guest devices can’t reach corporate servers, printers, management interfaces or each other.
- Controlled access: who connected, when and with which device is known; access is granted for a limited time.
- Resource limits: guest traffic doesn’t consume the bandwidth and wireless channel time of corporate applications.
- Logging: the records needed for legal obligations and incident investigation are kept, and personal data is collected in moderation.
Here we focus on the network architecture and configuration side. Security policy topics such as guest access policy, NAC integration and device posture checks are covered in the 802.1X and NAC migration plan article.
Layer 2 isolation: SSID, VLAN and client isolation
Isolation is built layer by layer; each layer closes the gap in the previous one. A design that relies on a single layer opens up at the first misconfiguration.
| Layer | Implementation | What does it block? |
|---|---|---|
| SSID | A guest-specific SSID | Sharing of corporate SSID credentials |
| VLAN | Separate VLAN and IP subnet | The same broadcast domain as corporate subnets |
| AP client isolation | Blocking traffic between clients on the same AP | Guests on the same AP reaching each other |
| Switch / gateway | Private VLAN, port isolation or ACL | Guests on different APs reaching each other |
| Firewall | Deny from the guest zone to internal networks | Access to the internal network via routing |
A mistake you may encounter: the guest VLAN is separate, but its gateway sits as an SVI on the core switch. On the core, all internal VLANs are already routed to each other; the guest VLAN falls into this table too. Separate VLAN, shared routing. The preferred approach is to terminate the guest VLAN’s gateway on the firewall or a separate router and not create an SVI for this VLAN on the core switch. The guest VLAN should be carried only on the trunks going to AP ports and the firewall.
If the gateway has to stay on an L3 switch, apply an ACL in the inbound direction that blocks internal address blocks. Cisco IOS/IOS XE example (guest subnet 10.90.0.0/22, gateway and DNS forwarder 10.90.0.1):
ip access-list extended GUEST-IN
permit udp any eq bootpc any eq bootps
permit udp 10.90.0.0 0.0.3.255 host 10.90.0.1 eq domain
permit tcp 10.90.0.0 0.0.3.255 host 10.90.0.1 eq domain
deny ip any 10.0.0.0 0.255.255.255
deny ip any 172.16.0.0 0.15.255.255
deny ip any 192.168.0.0 0.0.255.255
permit ip 10.90.0.0 0.0.3.255 any
!
interface Vlan90
description GUEST-WIFI
ip address 10.90.0.1 255.255.252.0
ip access-group GUEST-IN in
This ACL allows DHCP and the DNS service on the gateway, denies everything destined for private address blocks (RFC 1918), and allows the remaining traffic out to the internet. If the organization uses addresses outside RFC 1918 (e.g. a public IP block it owns), those blocks must also be added to the deny lines; a forgotten block is a door left open. After applying the ACL, try access from a guest device to internal servers, switch management IPs and internal DNS one by one. Don’t trust an untested ACL. For VLAN numbering and the subnet plan, see the VLAN design in enterprise networks article.
Internet egress: a separate line, DMZ or tunnel
Where guest traffic exits to the internet determines the strength of isolation and the operating effort:
- Separate internet line: the guest VLAN exits through a separate modem/firewall without touching the corporate network at all. Isolation is the clearest; suspicious traffic generated by guests isn’t associated with the organization’s public IP address. It requires the cost of an extra line and managing a second device.
- A separate zone on the firewall (DMZ-like): a single line is used, and the guest VLAN is defined as a separate interface or zone on the firewall. Everything from the zone to internal networks is denied, and only specified services are opened to the internet. It’s the most common and balanced option.
- Tunnel to the center (anchor): in multi-branch structures, guest traffic at the branches is carried through a tunnel to a controller/gateway at the center or in the DMZ and exits from there. Policy and logging are gathered at a single point; the cost is branch WAN bandwidth.
Whichever is chosen, send guest DNS queries not to internal DNS servers but to a forwarder on the gateway or to external resolvers. The internal namespace doesn’t leak out.
Captive portal authentication options
A captive portal requires the guest to authenticate or accept terms on a web page before going out to the internet. Choose the method according to the guest profile:
| Method | Suitable scenario | Point to watch |
|---|---|---|
| One-time code via SMS | Store, waiting room, high number of guests | SMS provider cost; a phone number is personal data |
| Sponsor approval | Corporate visitor, supplier | The approving employee needs to be reachable |
| Voucher (coupon code) | Reception, training, hotel-like use | A record of whom codes were given to should be kept |
| Terms of use acceptance | Low-risk areas with weak logging requirements | No credentials are collected; assess compliance with logging requirements |
Most portal complaints come from these three technical points:
- Operating system portal detection: phones and laptops detect the portal by querying a specific address after connecting. Don’t block this detection or the SMS/identity provider addresses in the pre-portal allow list (walled garden); if you do, the portal page never opens.
- HTTPS redirection: redirecting HTTPS requests to the portal page produces a certificate warning. The portal should have a valid certificate signed by a public CA; where possible, redirection should be done via HTTP detection requests. An expired portal certificate produces the same warning; put the renewal date on the calendar.
- Random MAC addresses: current versions of iOS, Android and Windows can use a random MAC address per SSID; in some versions this address changes over time. MAC-based “approve once, remember forever” setups may not work as expected; set the session duration accordingly.
Bandwidth limits and DHCP scope sizing
Limit bandwidth at two levels: per client (separate rate limits for download and upload) and for the guest network total (traffic shaping on the firewall or gateway). Setting only a per-client limit doesn’t prevent the corporate line from filling up when many guests connect at the same time. On the wireless side, set the guest SSID’s QoS class to the lowest priority (background or best effort); voice and video traffic keep priority in channel access.
The DHCP scope is calculated not by the number of guests but by the number of concurrent devices and the lease time. Worked example: at a facility, 400 different guest devices connect during the day, and at most 250 devices are present at the same time. If the lease time is set to 8 hours, the addresses of devices that have left stay reserved until the lease expires, and all 400 addresses may be needed; a /24 (254 usable addresses) isn’t enough. When the pool is full, a new guest connects, can’t get an IP and calls to say “Wi-Fi isn’t working.” When the lease time is reduced to 1 hour, the need approaches the number of concurrent devices. The rules we apply:
- Use a short lease time (30 minutes–2 hours) on the guest network.
- Choose the subnet to contain at least twice as many addresses as the expected number of concurrent devices (e.g. /23 or /22).
- Monitor DHCP pool usage and raise an alarm when a threshold is exceeded.
- Take into account that devices using random MACs may get more than one address on the same day.
Logging and record-keeping requirements
In Turkey, the obligations of organizations providing public internet access are regulated by Law No. 5651 and related regulations; in practice, securely keeping internal IP allocation records (which IP address was given to which device in which time interval) is assessed within this scope. Because details such as the retention period, record format and how integrity is ensured may change, they should be verified against the current texts on mevzuat.gov.tr and with a legal advisor.
Records recommended to be kept, technically:
- DHCP lease records: MAC address, assigned IP, start and end time.
- Captive portal authentication records: user identifier (phone, voucher code, sponsor), MAC and time.
- NAT records: which public IP and port the guest’s internal IP went out through.
- NTP synchronization on all devices for correct timestamps.
The last item is the one most often skipped. DHCP, portal and NAT records whose clocks don’t match can’t be linked to each other; the records exist but can’t be used. For collecting these records centrally and storing them tamper-proof, the approach in the central syslog server article can be applied. If personal data such as phone numbers is collected, a privacy notice, retention period and access authorization should also be defined under KVKK.
Guest network setup checklist
- The guest SSID is separate, the guest VLAN is separate, and it’s carried only on the necessary trunks.
- The guest gateway is on the firewall or a separate router; there’s no guest SVI on the core switch, or an inbound ACL has been applied.
- AP client isolation is enabled; isolation between APs is provided on the switch or gateway side.
- Access from the guest zone to internal subnets, management interfaces and internal DNS has been verified by testing.
- Per-client and total bandwidth limits are defined; the guest SSID is in a low QoS class.
- The DHCP lease time is short, the subnet size is at least twice the number of concurrent devices, and pool usage is monitored.
- The captive portal certificate is valid; the walled garden list contains the detection and identity provider addresses.
- DHCP, portal and NAT records are stored centrally and with correct timestamps; retention conditions have been verified against the legislation.
- The guest SSID broadcasts only on the necessary bands; turning it off after hours or a time schedule has been considered.
The guest network’s VLAN and routing design can be handled together under the VLAN, routing and network segmentation service, and its SSID, portal and RF configuration under the wireless network installation and optimization service.
Frequently asked questions
Is a separate internet line required for guest Wi-Fi?
It isn’t required, but it provides the clearest isolation and prevents guest traffic from affecting the corporate line’s capacity and the reputation of the public IP address. If a single line is used, the guest VLAN should be placed in a separate zone on the firewall, access to internal networks should be explicitly blocked, and guest traffic should be limited by a bandwidth policy.
Is it enough to enable client isolation only on the AP?
No. Client isolation on the AP usually blocks traffic only between clients connected to the same AP. To prevent guests connected to different APs from reaching each other, private VLANs, port isolation or an ACL at the gateway on the switch side, or an architecture that tunnels traffic to a central point, is required.
Which records need to be kept on the guest network, and for how long?
In Turkey, the obligations of organizations providing public internet access are set by Law No. 5651 and related regulations. Clarify which record is to be kept, in what form and for how long based on the current provisions in the legislation and together with a legal advisor; separately assess KVKK obligations for data collected, such as phone numbers.
Should encryption be used on the guest SSID?
On open guest networks, over-the-air traffic isn’t encrypted. On clients that support it, OWE (Wi-Fi Enhanced Open) provides per-connection encryption without asking for a password. For devices that don’t support OWE, transition mode or a shared password (WPA2/WPA3-Personal) can be used; a captive portal is for authentication and doesn’t provide encryption.