Before opening the firewall catalogue and creating twelve random VLANs, I look at what the network already lets through. No need to draw every cable. I care about paths that go far too wide and dependencies that can drag everything else down with them.
Reading the current architecture
The MPLS network connects sites and the datacenter. It is a transport mechanism, not an automatic security boundary. Two devices on the same wide-area network should be allowed to communicate only when an explicit rule supports a real need.
The question is dead simple: how far can a compromised account, a bad rule or an infected contractor laptop go? If the answer is “nearly everywhere”, we have found the problem.
1. A wide-area network mistaken for a trust zone
Site-to-site connectivity makes operations easier, but it also creates paths from one site to another. Safety equipment at one location does not need to discover or contact equivalent equipment everywhere else.
The first question is therefore not “which firewall should we buy?” but “which inter-site exchanges are actually required?”. Any flow without an owner, justification, or operational need becomes a candidate for removal.
2. A flat safety LAN inside each site
Cameras, access controllers, and related devices historically share one VLAN. This has two consequences: address space is running out, and a compromise can reach devices that perform very different functions.
Segmentation means more than creating extra VLANs. Each segment needs a policy: who may initiate traffic, towards which destination, on which ports, and how denied traffic will be logged.
3. Contractors placed too close to the core
External operators need maintenance access, not a general presence on the MPLS network. The current path makes three simple questions hard to answer: which contractor connected, which equipment did they reach, and when did that access end?
- External devicea separate trust level
- MPLS accessreach is too broad
- Other sitesunnecessary paths
- Safety devicepossible operational impact
The target design needs an identifiable, authenticated, and logged entry point, followed by restrictions to a defined list of devices or services. Least privilege then becomes a property we can verify rather than an intention written in a document.
4. Central services that shape the migration
DHCP, DNS, and Active Directory currently live in the external datacenter. Moving them to two pilot sites cannot be treated as a simple virtual-machine relocation. It changes dependencies, name resolution, address allocation, and failure scenarios.
The normal path must be documented first. The loss of one pilot site, the MPLS link, and the datacenter should then be tested separately. Redundancy on paper only becomes resilience once degraded modes have been observed.
What comes next
The next module turns this mess into trust zones, filtering rules and a proper contractor path. The thread stays the same. Every access route should reach less without breaking what production genuinely needs.