Skip to content

Network Segmentation Best Practices (Classic + Modern Guide)

Network Segmentation Best Practices

Ask three network engineers to define "network segmentation" and you'll likely get three different scopes — one thinking VLANs, one thinking zero trust, one thinking cloud security groups. They're not wrong, exactly. They're describing different points on the same spectrum, and most segmentation projects fail not because the technology is wrong, but because nobody agreed on which point they were actually building toward.

The network segmentation spectrum A horizontal diagram showing four stages from left to right: flat network, VLAN and subnet segmentation, DMZ and zone-based segmentation, and microsegmentation with zero trust, illustrating that these are points on a spectrum of granularity rather than competing single choices. Flat network VLAN / subnet segmentation Zone-based / DMZ architecture Microsegmentation / zero trust Increasing granularity, increasing operational overhead →
These approaches aren't competitors — most mature networks use several of them layered together. Source: Generated by UnpanicTech.

Classic segmentation still does most of the work

It's tempting to treat VLANs and DMZs as legacy technology now that "zero trust" dominates the conversation. They aren't. CISA's segmentation guidance still describes physical and virtual zone separation — dividing a network into subnetworks that each act as their own security boundary — as an effective way to reduce risk, particularly for separating IT and operational technology (OT) environments [1]. A properly configured DMZ, sitting between the internet and internal networks with a firewall on each side, still does real work: it stops a compromised public-facing service from becoming a direct path into internal systems, and it gives defenders a natural chokepoint to monitor [1].

The classic model — internet, DMZ, internal IT network, OT/control network, each separated by a firewall — maps closely to the Purdue reference architecture used across industrial environments. It's not glamorous, but it's the foundation almost everything else in this article sits on top of.

Classic zone-based segmentation architecture A left-to-right diagram showing the internet connected through a firewall to a DMZ, then through a second firewall to an IT business network, then through a third firewall to an OT network, illustrating layered zone separation. Internet DMZ IT / Business network OT / Control network FW FW FW
Each boundary is a firewall and a monitoring point, not just a routing decision — a layout consistent with CISA's segmentation guidance for IT/OT separation. Source: Generated by UnpanicTech.

Where zero trust and microsegmentation actually change things

NIST SP 800-207 describes the core shift plainly: stop granting trust based on network location, and instead verify every request based on identity, device posture, and context, regardless of whether it originates "inside" or "outside" the perimeter [2]. That's a meaningful departure from VLAN-based thinking, where being on the right subnet is treated as sufficient proof you belong there.

CISA's more recent guidance on the path to microsegmentation frames this as a maturity progression rather than a rip-and-replace: organizations typically move from broad perimeter-based segmentation toward finer-grained, workload-level boundaries as their zero trust program matures, because microsegmentation lets policy travel with the workload instead of being tied to a fixed network location [3]. This piece isn't re-running a specific microsegmentation product comparison — the point here is where microsegmentation fits in the bigger picture, not which vendor implements it best.

Cloud environments complicate this further. NSA and CISA's joint guidance on cloud segmentation notes that on-prem networks traditionally relied on specialized appliances to achieve zero trust, while cloud environments can tie identity directly into network requests, encrypt traffic end-to-end by default, and apply microsegmentation through native constructs like security groups rather than physical network boundaries [4]. A VPC security group isn't a VLAN with a different name — it's an identity-and-policy construct that happens to also filter traffic.

Decision tree: which approach fits your environment

There's no universally "best" segmentation model — the right starting point depends heavily on what you're actually segmenting:

  • Legacy OT / industrial environment: Start with classic zone-based segmentation — IT/OT separation, a properly monitored DMZ, and firewalls at each boundary. Microsegmentation at the OT layer is possible but often constrained by device age and protocol support; don't force it before the basics are solid.
  • Traditional on-prem enterprise network: VLANs and subnetting remain the practical backbone. Layer zero trust principles (identity-based access, device posture checks) on top for sensitive segments rather than replacing the underlying network design outright.
  • Cloud-native or hybrid environment: Lean on native identity-and-network constructs — security groups, VPC boundaries, service mesh policies — informed by zero trust principles from the start, since retrofitting VLAN-style thinking onto cloud infrastructure tends to fight the platform.
  • Regulated or high-value asset environments (payment systems, PHI, ICS): Combine both — classic zone isolation for the coarse boundary, microsegmentation for the specific systems that need it, driven by what the compliance framework or risk assessment actually requires, not by what's trendy.
Segmentation approach decision tree A flowchart starting from a question about environment type, branching to four paths: legacy OT leading to zone-based IT/OT separation, on-prem enterprise leading to VLANs plus targeted zero trust, cloud-native leading to native security groups and identity-based policy, and regulated high-value assets leading to a combination of zone isolation and microsegmentation. What's the environment? Legacy OT → Zone-based IT/OT separation On-prem enterprise → VLANs + targeted zero trust Cloud-native → Native security groups + identity Regulated / high-value → Zone isolation + microsegmentation
A starting point, not a rulebook — most real environments combine two or more branches. Source: Generated by UnpanicTech.

Where segmentation projects actually fail

The technology is rarely the reason a segmentation project stalls or quietly becomes ineffective. It's usually one of these:

  • No accurate asset inventory. You can't segment what you don't know exists. CIS Control 13 assumes traffic filtering between segments is backed by an understanding of what's actually talking to what — without that, rules get written from guesswork and either break things or leave gaps [5].
  • "VLANs in name only." Segments exist on paper, but routing between them is wide open, or a single overly broad firewall rule effectively flattens the design back to one network.
  • Unmonitored east-west traffic. Perimeter monitoring gets attention; traffic between internal segments doesn't. CIS Control 13 explicitly calls out traffic filtering and flow log collection between network segments as a distinct safeguard from perimeter defense — it's easy to build the boundary and forget to watch what crosses it [5].
  • Firewall rules that never get reviewed. Rules accumulate for years, nobody owns cleanup, and "temporary" exceptions from a project three years ago are still open.
  • Treating microsegmentation as a product purchase. CISA's own guidance frames the move to microsegmentation as an architectural and organizational shift requiring leadership buy-in, not just a tool deployment [3]. Buying a microsegmentation platform without the operational process to maintain policies tends to produce shelfware.
Common segmentation failure modes A conceptual, non-statistical diagram showing four categories of failure: no asset inventory, VLANs in name only, unmonitored east-west traffic, and stale firewall rules. No asset inventory Rules written from guesswork VLANs in name only Routing left wide open Unmonitored east-west traffic Boundary built, not watched Stale firewall rules No owner, no review cadence
Conceptual illustration of recurring failure patterns — not a measured breakdown of any specific organization's incidents. Source: Generated by UnpanicTech.

A practical starting checklist

  • Build (or validate) an asset inventory before writing a single new segmentation rule.
  • Document the current network architecture as an actual diagram, not tribal knowledge — CIS Control 12 treats this as a standing requirement, not a one-time exercise.
  • Confirm your DMZ actually isolates — test that a compromised DMZ host can't reach internal systems directly.
  • Add flow logging and filtering between existing segments before investing in finer-grained microsegmentation.
  • For cloud workloads, default to identity-tied security groups rather than replicating on-prem VLAN logic.
  • Assign an owner and a review cadence for firewall/segmentation rules — "temporary" rules are where drift starts.

Sources & References

  1. Layering Network Security Through Segmentation — CISA
  2. NIST SP 800-207, Zero Trust Architecture — NIST
  3. The Journey to Zero Trust: Microsegmentation in Zero Trust — Part One — CISA
  4. Implement Network Segmentation and Encryption in Cloud Environments — NSA & CISA
  5. CIS Critical Security Control 13: Network Monitoring and Defense — Center for Internet Security
Zero Trust Architecture & Subnet Planning Resources

When carving out isolated VLANs and administrative jump zones, verify your netmasks with our CIDR subnet calculator and lock down administrative entry points using our bastion host hardening checklist.

NK

Cybersecurity Researcher & Technical Editor

UnpanicTech publishes independent cybersecurity research and defensive security analysis. Naseem Khan serves as Technical Editor, with a focus on vulnerability analysis, defensive security, incident response, cloud security, and practical security engineering.

Leave a Comment (Authenticated Users)