A single compromised laptop on a flat /24 subnet can probe every neighbour it can reach, and in many networks nothing stops it. We keep finding this exact design during network assessments, and it still turns up far more often than most engineers expect. Microsegmentation was built to close that gap. Understanding how it works is the first real step toward fixing the problem, so we want to walk through it the way we explain it to our own engineers.
Put simply, microsegmentation applies fine-grained policy between individual workloads. Rather than trusting every host that happens to share a VLAN, traffic between systems is denied by default. Only the flows someone has explicitly approved get through. The enforcement point sits close to each workload, and the rules reference identity or attributes such as application and environment instead of just subnet boundaries. That shift is where most of the value comes from.
A picture we often use with clients is moving the security checkpoint from the building’s front door into every single room. A visitor who gets past the lobby still has to show a badge at each door they try to open.
Why Flat Networks and Coarse VLANs Fail Against Lateral Movement
Most intrusions we have investigated followed the same pattern. One host falls, and then it reaches its neighbours simply because they live in the same broadcast domain. The attacker rarely needs a clever exploit for that second step. Connectivity does the heavy lifting. Once an attacker has a foothold, the network itself becomes their map.
VLANs group machines according to where they sit on the network, not according to what they do. That distinction matters a great deal. A VLAN cannot express a rule like “the web tier may reach the app tier on port 8443, and nothing else.” It can only state that these machines belong together, and everything in that group is trusted equally. Once you accept that, the weakness is obvious.
Consider a flat 10.20.0.0/24 holding web servers, application servers, and a reporting database. A compromised web host runs a quick sweep, spots 10.20.0.31 listening on 5432, and connects without resistance. Nothing on the switch was misconfigured in the sense of a typo. The design simply had no vocabulary for saying no between those hosts. That missing vocabulary is exactly what microsegmentation supplies.
The Building Blocks: Workloads, Labels and Policy Objects
A workload is any endpoint that needs its own rule set. That might be a virtual machine, a container, or a bare-metal server. Each one receives labels such as application, environment, and data classification. Those labels become the grouping key for policy, and getting them right early saves a lot of pain later.
The payoff is that you stop writing rules against IP addresses. Addresses change whenever a segment is re-addressed or a service migrates. Rules tied to them decay quietly until nobody trusts the rule base anymore. Labels survive those changes because they describe what a system is, not where it happens to sit.
Policy then reads as a plain statement: an allowed source group, a destination group, a protocol, and a port, with default deny underneath everything. Written this way, the rule set stays readable to someone who did not author it. Most long-lived firewall rule bases cannot claim that.
Host-Based or Network-Based: Where the Rule Actually Runs
Enforcement happens in two broad places. Host-based enforcement runs an agent or a kernel filter on the endpoint. On Linux that could be nftables, iptables, or eBPF-based tooling. The main advantage is that the rule travels with the workload, even when it moves to another subnet or to a cloud provider.
Network-based enforcement happens on switches, routers, or firewalls as traffic crosses them. No agent is needed on each system, which makes it easier to deploy on legacy hardware that cannot run one. The trade-off is that policy is bound to the network path. A workload that moves can slip past rules written for its old location, and traffic that never crosses your equipment is invisible to the policy.
In our experience, many teams end up using both. Host-based enforcement covers the servers they control directly, while network-based controls guard the choke points between zones. The right mix depends on how much of your estate you can instrument on the endpoint, and that answer is rarely as clean as the architecture diagram suggests.
Building a Policy Model from Flow Data
Do not write rules from memory. Collect flow records, whether NetFlow, IPFIX, sFlow, or cloud flow logs, over a baseline period. Two weeks is a reasonable minimum, because it captures month-end jobs and weekly backups that a shorter window would miss entirely. We have seen teams build confident rule sets from three days of data and then break a backup job the first Friday after go-live.
Group the talkers by label and see which group pairs actually communicate, along with the ports they use. Each observed pair becomes a candidate allow rule. Anything absent from the baseline deserves questioning before it earns a rule. Before enforcing, review rules that match nothing or look overly broad, since mistakes tend to hide there.
Here is a small nftables example for a default-deny web-to-app flow on port 8443:
table inet seg {
chain forward {
type filter hook forward priority 0; policy drop;
ip saddr 10.20.1.0/27 ip daddr 10.20.2.0/27 tcp dport 8443 accept
}
}
Notice that the default policy is drop and the only accept corresponds to the flow you observed. Everything else, including traffic nobody remembered to document, is refused until someone makes a deliberate decision to allow it. That deliberate step is the discipline the whole model depends on.
Where Microsegmentation Beats VLAN Segmentation
VLANs remain useful for coarse separation, such as keeping guest Wi-Fi apart from corporate networks. Microsegmentation does not replace them. It answers a finer question: which host may talk to which host, and on which port.
Take a shared VLAN that holds payment servers alongside reporting servers. Under pure VLAN rules, both sit in one broadcast domain, and reporting can reach payment systems freely. With microsegmentation, you block reporting-to-payment access without renumbering a single machine. The change is a policy edit instead of a network project, which is why it tends to happen far sooner in practice.
Rolling Out Without Breaking Production
Start in monitor or audit mode. Log every flow that would have been denied, study the results carefully, and then move to enforcement one application group at a time. Keep a rollback path and a documented break-glass procedure before touching any production host. Rollouts that skip these steps are the ones that end up in post-incident reviews.
Before each group goes live, confirm that the scope covers only one application group and that the baseline spans at least two weeks. Agree on the label standard, get the change approved, watch the would-be denials in audit mode, and test the rollback before you ever need it. A rehearsed rollback is worth more than an elegant policy you cannot undo.
Mistakes We See Again and Again
Labelling by IP address is the first and most common error. It works until someone re-addresses a subnet, at which point the policy quietly stops matching anything. Another is writing allow-any rules during the learning phase and never tightening them afterwards. That leaves you with a segmentation project that segments nothing in practice.
The most painful mistake is forgetting the unglamorous traffic. DNS, NTP, patch servers, and monitoring agents all need paths. Turn on default deny without them and you will have an outage before lunch. Management and backup networks belong in the policy as well, and they are often the first things teams overlook.
Cost, Effort and Where to Start
Cost depends on the licensing model, the number of workloads, and how much your team already knows. Quotes vary widely between vendors and between environments, so ask for pricing against your own estate rather than trusting a figure from any blog post, this one included.
The real effort is ongoing: time spent on flow analysis, label governance, and regular rule review. We suggest starting with one high-value zone, usually the database tier. Prove the approach there, then expand outward from that foothold. A small, successful first zone builds the operational confidence the rest of the project depends on.
How This Fits With Zero Trust
Microsegmentation is one of the practical building blocks of a zero trust model, because it stops assuming that anything inside the network is safe. If you want the broader argument for why the network edge alone is not enough, read our piece on The Zero Trust Perimeter Myth.
For readers who want to see how these controls sit alongside other infrastructure work, the cybersecurity category covers related topics such as lateral movement detection and identity-aware access. The MONA AI team has also written about how automated analysis can help teams interpret large volumes of flow telemetry, which pairs well with the baseline work described above.
Frequently Asked Questions
Is microsegmentation the same as zero trust?
No. Zero trust is the overarching principle of verifying every access request. Microsegmentation is one technique that enforces that principle between workloads.
Do I need to replace my VLANs to use microsegmentation?
No. Keep VLANs for coarse separation and layer microsegmentation on top for the finer rules.
Can microsegmentation work without installing agents on every server?
Yes, network-based enforcement can do it on switches and firewalls. The trade-off is that policy follows the network path, so it may miss traffic that never crosses your devices.
How long does it take to build a usable policy from flow data?
Plan for a baseline of at least two weeks plus review time. Complex environments take longer, so budget for iteration.
What is the first application I should segment?
Start with the database tier or another high-value zone where the blast radius of a compromise is largest and where the flows are relatively well understood.