On this project, the obvious move would be to split the old LAN into several VLANs and call the job done. That is also a good way to make the network harder to operate without making it much safer.
The split is only the visible part. I care about the rule behind it. Who talks to whom, which side starts the connection and what breaks if I block it? Until I have those answers, picking a VLAN number achieves very little.
- Inventorydevices, roles and dependencies
- Zonestrust level and function
- Trafficsource, destination, service and owner
- Testsallowed, denied and logged
The decision comes before the VLAN
A VLAN does separate layer-two broadcast domains. Once a router or firewall moves traffic between them, the policy at that crossing is what matters. Allow everything between every VLAN and all we have really organised is the broadcasts.
I think in trust zones instead. Cameras do not have the same job as admin workstations. Infrastructure services should not be exposed like the entry point used by contractors. That separation makes sense because their use and risk differ, not because the devices have a different badge on the front.
It is easy to go too far as well. A zone per building, camera model or passing thought becomes painful to run. If two groups need the same policy and the same traffic, separating them probably buys very little.
Write down the need before the rule
Before touching the firewall, I put the exchanges in a table. It does not need to be a forty-page document. One row should explain the path and point to the person who knows why it exists.
| Source | Destination | Service | Initiator | Owner | Reason | Logging |
|---|---|---|---|---|---|---|
| camera zone | video server | required stream | camera | safety team | display and recording | denies + anomalies |
| admin workstation | managed devices | administration | admin workstation | operations | maintenance | allows + denies |
| device | internal DNS | resolution | device | infrastructure | internal names | errors |
The values above are deliberately generic. I am not going to invent a camera’s ports from the first product sheet I find online. I check the device, its documentation and the captures. Then I find somebody willing to own the need. A rule with no owner almost always stays open “just in case”.
Watch the traffic without trusting it
I still want to deny inter-zone traffic that has no reason to exist. I just do not switch the whole network to blocking overnight.
First I observe long enough to see more than one quiet Tuesday morning. Then I compare the captures with the documentation and with what the teams say. This is usually where an old relay, a monthly backup or some hard-coded device turns up.
A capture is not a ready-made allow list. It only shows what happened during that window. Useless traffic appears perfectly well. A rare but important job may be missing altogether.
Keep an address plan we can still read in six months
I want to recognise the site and function of an address without opening an archaeological spreadsheet. Contiguous blocks per site make routing and ACLs easier. Reusing the same zone logic across sites also means we do not have to relearn the network during every intervention.
I leave room. Filling every subnet to 95% during the migration only guarantees another migration later. Size comes from the real device count, expected growth and acceptable layer-two noise. “Everything is a /24 because it always has been” is not a reason.
The plan records at least the networks, gateways, DHCP relays and reservations. I avoid overlapping addresses between sites too. Those duplicates always reappear at the worst time, usually when a new tool or tunnel has to be connected.
Put the filter on the path packets actually take
A rule is useless if routing lets traffic go around it. I check the forward path, return path, real gateway and what changes during failover. A switch ACL may be enough for some local exchanges. A firewall gives us more state and better logs. The right choice mostly depends on where the packet really travels.
For acceptance, I start with one specific flow. The camera may need to reach the video server, but it should not start a connection to the admin workstation. I test both from the correct network and find both events in the logs. If the useful flow passes, the other one stops and the application still behaves normally, the rule is starting to earn some trust.
I keep rollback close by too. Finding a missed dependency during a pilot is fine. Finding one with no way back gets ugly very quickly.
Only then does the diagram begin to mean something. It is no longer a set of separate boxes. It summarises paths we have actually tried.
If you are planning to separate guest devices from business systems, see Bacchus Security’s network design service.