Campus Segmentation Design: From VLANs to Unified Fabric
Campus segmentation often starts with virtual local area networks (VLANs) because VLANs are familiar and easy to create. The problem is that VLANs alone do not provide a complete security model. They identify broadcast domains. They do not automatically express identity, role, intent, or least privilege.
Design takeaway: use VLANs for attachment, virtual routing and forwarding instances (VRFs) or virtual networks for macro boundaries, and Security Group Tags (SGTs) for role-based microsegmentation policy.
The VLAN Sprawl Problem
Review each existing VLAN and ACL before migrating it. Record whether it provides attachment, an enforced security boundary or a temporary exception, and identify its owner. That distinction determines which controls must survive the migration.
A better design separates attachment from policy. A device can attach through a VLAN, but its access should be determined by identity, role, segment, and destination.
Macrosegmentation
Cisco SD-Access guidance describes macrosegmentation with virtual networks or VRFs. Use this for big boundaries such as Guest, Internet of Things (IoT), Corporate, operational technology (OT), Payment Card Industry (PCI), and Shared Services.
The best practice is restraint. Keep the number of macro segments small, because every virtual network (VN) or virtual routing and forwarding (VRF) affects routing, border handoffs, route leaking, firewall policy, monitoring, and operations.
Microsegmentation
Microsegmentation handles the fine-grained question: inside a broader domain, which roles or device groups can talk to each other? This is where SGTs and group-based policy are useful.
- Employees may reach business apps but not building automation.
- Cameras may reach video management systems but not user subnets.
- Printers may receive print traffic but not initiate broad lateral access.
- Admins may reach management interfaces only from hardened workstations.
Implementation Order
Observe the current permitted flows and define a small pilot without removing existing denies. Record required dependencies and forbidden paths, then stage only the approved policy changes. Observation is not a reason to open access that is already blocked.
- Inventory users, devices, applications, and shared services.
- Define macro segments and inter-segment routing paths.
- Define SGTs for meaningful roles and device classes.
- Monitor policy hits and unexpected traffic.
- Where the existing pilot segment is permissive, introduce approved targeted denies and progress toward tested least privilege; retain existing restrictions elsewhere.
A segmentation design succeeds when security can explain the policy and network operations can troubleshoot it at 2 a.m.
Design Detail: Segmentation Has a Cost Curve
VRFs require routing, border handoff and shared-service review. SGTs require classification, propagation and enforcement validation. VLANs require addressing, DHCP and gateway ownership. Choose controls that meet the risk requirement and that the team can support and test, not simply the lowest-cost option.
The migration should begin by reducing ambiguity. Label existing VLANs as attachment-only, security-boundary, legacy-exception, or retirement candidates. Then decide which boundaries deserve VRFs and which roles deserve SGTs.
Implementation Details
- Do not use a new VRF when a Security Group Tag (SGT) policy inside an existing VN is enough.
- Do not use an SGT when the endpoint cannot be reliably classified.
- Avoid hundreds of groups unless the operations team can maintain the matrix.
- Use traffic observation before default-deny enforcement.
- Document shared-services paths and exception ownership.
Segmentation Control Model
The clean model is layered. VLANs attach endpoints. VRFs or virtual networks create macro boundaries. SGTs express role and device intent. security group access control lists (SGACLs) enforce the role matrix. Firewalls still belong at places where inspection, network address translation (NAT), threat prevention, or deep logging is required.
| Control | Use It For | Do Not Use It For | Evidence |
|---|---|---|---|
| VLAN | Layer 2 attachment, DHCP scope, wireless or wired onboarding behavior | Long-term security intent by itself | Switchport assignment, wireless local area network (WLAN) mapping, DHCP lease, endpoint identity. |
| VRF or virtual network | Large trust boundaries such as Guest, Corporate, OT, PCI, IoT, and Shared Services | Every department or every small exception | Route table, border handoff, route leak policy, firewall path. |
| SGT | Role, device class, or identity-driven policy inside a macro segment | Endpoints that cannot be classified reliably | NAC authorization, SGT assignment, propagation state, endpoint session. |
| Security Group Access Control List (SGACL) | East-west allow and deny decisions by source group and destination group | Deep application inspection or user activity auditing | Policy matrix version, hit counts, deny logs, switch enforcement point. |
| Firewall | North-south inspection, inter-VRF policy, internet edge, regulated logging | Compensating for an unreadable campus policy matrix | Rule ID, session log, threat log, change ticket. |
SGT and SGACL Example
This is an illustrative policy matrix, not configuration. Keep the first enforcement wave small enough to review each required service, enforcement point and denied path with its owner. Do not infer platform support or complete application dependencies from these examples.
| Source SGT | Destination SGT | Allowed Services | Default | Operational Note |
|---|---|---|---|---|
| Employee | Business_App | Hypertext Transfer Protocol Secure (HTTPS), app-specific Transmission Control Protocol (TCP) ports | Deny all else | Validate app dependencies before blocking ephemeral paths. |
| Employee | Print_Server | Approved print service, such as Internet Printing Protocol (IPP) | Deny direct printer admin | Review the separate print-server-to-printer rule; this row does not grant direct printer access. |
| Camera | Video_Management | Vendor-required ports from camera to recorder | Deny all else | No camera-initiated access to user VLANs. |
| IoT | DNS_NTP_Update | Domain Name System (DNS), Network Time Protocol (NTP), vendor update endpoints | Deny lateral access | Use firewall or proxy where vendor destinations need inspection. |
| Admin_Workstation | Network_Management | Secure Shell (SSH), HTTPS, Remote Desktop Protocol (RDP) to jump hosts, Simple Network Management Protocol (SNMP) from tools | Deny from non-admin groups | Require hardened workstation posture and phishing-resistant multi-factor authentication (MFA). |
| Guest | Any_Internal | None | Deny | Guest should egress through internet controls only. |
Enforcement Safety
Check classification errors, hidden dependencies and unsupported enforcement behavior before expanding the pilot. The stages below each name the evidence needed and a condition that stops the rollout.
| Stage | Goal | Required Evidence | Stop Condition |
|---|---|---|---|
| Observe | Learn real flows before denying traffic | NetFlow or telemetry for source, destination, port, app, user, device, SGT, and business owner | Unknown critical app paths or unclassified endpoint classes. |
| Classify | Assign SGTs consistently | NAC authorization logs, endpoint profile confidence, fallback handling | Large number of endpoints landing in Unknown or default groups. |
| Monitor policy | Simulate or log denies where supported | Policy hit counts, unexpected destination list, owner review | High-volume denies to shared services, identity, DNS, DHCP, or update systems. |
| Partial enforce | Block narrow high-risk lateral paths | Positive tests for required apps and negative tests for blocked movement | Help desk tickets without policy evidence or rollback clarity. |
| Default deny | Move mature groups to least privilege | Exception register, expiry dates, change approval, rollback tested | Unowned exceptions or undocumented route leaks. |
Pilot One Building and One User Group
A useful first pilot is deliberately narrow: one building, one employee group, one shared service, and one low-risk device class such as printers. Record the existing VLANs, VRFs, gateway paths, DHCP scopes, DNS dependencies, identity sources, access-control lists, and help-desk baseline before changing enforcement. That snapshot is the rollback target, not merely background documentation.
- Place a small representative endpoint set in the pilot and confirm it receives the expected identity and segment before applying a deny.
- Test required business flows from the endpoint, then run explicit negative tests toward management, camera, guest, and unrelated application segments.
- Move one endpoint, expire one identity session, and simulate an unavailable identity service so the team sees reauthentication and fallback behavior.
- Trace a permitted and denied flow through endpoint classification, policy resolution, propagation, enforcement, and the SIEM record.
- Restore the original VLAN, gateway, and access policy from the recorded baseline. Do not expand until operations can perform that rollback inside the approved change window.
The pilot should stop when a critical dependency has no owner, an endpoint lands in a permissive unknown group, a deny cannot be explained from logs, or the old path cannot be restored predictably. These are design failures to resolve, not reasons to add permanent broad exceptions.
Evidence, Security Information and Event Management (SIEM), and Incident Workflow
| Evidence Source | Fields to Normalize | How the security operations center (SOC) uses it |
|---|---|---|
| NAC or identity platform | User, device, profile, authorization rule, assigned SGT, posture result | Confirm whether a blocked flow is a real policy decision or a classification failure. |
| Switch or fabric enforcement | Source SGT, destination SGT, SGACL rule, permit or deny, enforcement node | Detect lateral movement attempts and identify which device enforced the deny. |
| Flow telemetry | Source, destination, port, protocol, bytes, app, site, segment | Build allow lists, find hidden dependencies, and hunt for new east-west paths. |
| Firewall logs | VRF, zone, rule ID, threat verdict, user, app, uniform resource locator (URL), NAT | Correlate inter-segment movement with inspection results and threat events. |
| Change and exception register | Policy version, requester, approver, business reason, expiry | Let security orchestration, automation, and response (SOAR) auto-open review tasks for expiring or abused exceptions. |
What This Does Not Protect or Validate
- Segmentation does not confirm an endpoint is clean. It only limits what that endpoint should be able to reach.
- An SGT-based policy does not protect traffic if endpoint classification is wrong, missing, or easy to spoof.
- A deny in the campus does not replace vulnerability management for applications, printers, cameras, controllers, or network devices.
- VRFs reduce blast radius, but route leaking and shared services can quietly rejoin separated worlds if they are not reviewed.
- A policy matrix is not production-ready until operations can explain a blocked flow using identity, classification, propagation, and enforcement evidence.
Evidence boundary: the policy examples and rollout gates are documentation-backed design guidance. They are not results from a TechGeeks Catalyst Center, ISE, TrustSec, or EVPN lab, and they do not establish scale or interoperability for a specific software train.
Cisco References
- Cisco SD-Access Design Guide
- Cisco SD-Access Macro Segmentation Guide
- Cisco Ethernet virtual private network (EVPN) microsegmentation
- Trust at machine speed
Related foundation post: Cisco Live 2026: Network Announcements That Matter.



One thought on “Campus Segmentation Design: From VLANs to Unified Fabric”