Monitoring Traffic Between Zones
If you followed the trust zones guide, you now own a zone-pair matrix. That is a grid that says, for every pair of zones, what traffic is allowed, what is merely tolerated and logged, and what is blocked. That grid is your intended state. This guide is about the other half of the picture: what is actually crossing between your zones, right now, and how you would know.
Here is the honest starting point for most small networks: nobody is looking. Traffic between VLANs passes through the core switch. The core switch routes it in silicon at wire speed and tells you almost nothing about who talked to whom. The internet firewall logs the gate; the inside of the building has no cameras. That is the default state of nearly every network that grew organically, not a failing on your part.
By the end you will know the four places inter-zone traffic can be observed and the limits of each. You will know what to log for every zone pair and how to build a baseline per pair. You will know how to turn your matrix into a living dashboard that shows new pairs and new ports the day they appear.
In this guide
- The observation points: where inter-zone traffic must pass
- The honest limit: traffic that never crosses a chokepoint
- What to log for every zone pair
- Turning addresses into zones
- Building a baseline per zone pair
- Alerting on new pairs and new ports
- The matrix as a living dashboard
- Walkthrough: from raw flows to one dashboard cell
- Where to go next
The observation points: where inter-zone traffic must pass
One fact makes inter-zone monitoring possible at all. Two zones live on different subnets, and traffic between different subnets has to be routed. That means handing it to a device that knows the way from one subnet to the other. That device is a chokepoint: every inter-zone conversation passes through it, so watching it means watching all of them. In practice there are four kinds, and most networks have at least two.
1. The inter-VLAN routing point
In most small networks this is the core switch — a layer-3 switch that forwards frames within a VLAN and routes packets between VLANs. Or it is a router doing the same job. Everything between your user VLAN and your server VLAN goes through it. What it tells you depends on the model. Almost all expose per-VLAN interface counters over SNMP, so you can see how much traffic enters and leaves each zone. Many can log hits on access control lists, a crude "this rule matched" record. Fewer can export flow records. A core switch that routes in hardware is fast precisely because it does not stop to think about each conversation. And that is why it is usually a thin source of detail.
2. A firewall between zones
If any of your zone boundaries pass through a firewall, you have the richest possible view. It could be a dedicated internal firewall, or the internet firewall with internal VLANs attached to it. A firewall is stateful: it tracks each session from first packet to last. So it can log who started it, where it went, which port, how many bytes, and how long it lasted. It also knows which rule allowed or denied the session. The catch is that most firewalls log only denies by default. To see the allowed inter-zone traffic you must switch on session logging for the allow rules at the boundary. Do it anyway; the allowed traffic is where the story is.
3. Flow export at the core
A flow record is a one-line summary of a conversation: source address, destination address, protocol, source and destination ports, bytes, packets, start time, end time. No payload. Flow export is a feature of routers, layer-3 switches, and firewalls that sends these summaries to a flow collector, which stores and indexes them. Flow formats differ in details — some sample one packet in every so many rather than counting all — but the shape of the data is the same. Flow export is the workhorse of inter-zone monitoring. It sees every routed conversation and is cheap to store for weeks. It answers "who talked to whom, on what port, how much" without the weight of full packet capture. If your core device supports it, turn it on for the VLAN interfaces at your zone boundaries and point it at a collector. If it does not, that alone can justify a flow-capable device at the boundary that matters most.
4. Mirror ports at chokepoints
A mirror port (also called a SPAN port) tells a switch to copy every packet crossing one port onto another, where a sensor listens. Mirror the uplink between your access switches and the core, or the link to the boundary firewall. Then the sensor sees the complete packets of every inter-zone conversation that passes there. It is the most detailed view and the most expensive to keep. Use flows for the everyday view and mirrors when a flow tells you to look closer.
| Observation point | What you see | What you do not see | Effort |
|---|---|---|---|
| Core switch or router (interface counters, ACL logs) | Bytes in and out per VLAN; which ACL entries are being hit | Individual conversations; who talked to whom | Low: SNMP polling you probably already do |
| Firewall at a zone boundary | Every session: source, destination, port, bytes, duration, rule; allows and denies | Traffic on boundaries the firewall does not sit on | Medium: enable session logging on allow rules, ship logs to one place |
| Flow export at the core | Every routed conversation as a summary record; weeks of history | Payload; same-VLAN traffic; sampled formats may miss short conversations | Medium: enable export on the device, run a collector |
| Mirror port to a sensor | Full packets on the mirrored link | Anything not on that link; storage runs out quickly | High: sensor hardware, disk, and someone to analyze it |
The honest limit: traffic that never crosses a chokepoint
Now the limit, because it matters for the highest-value alert in this whole topic. Two hosts on the same VLAN do not need a router to talk. The frames go from one switch port to another and never touch the core's routing engine, never pass a firewall, and never generate a flow record at the core. Two workstations in the user zone reaching each other on an administrative port — the classic sign of something spreading — is exactly this kind of traffic. Your chokepoints are blind to it.
You have three realistic options, in rising order of effort. First, prevent it. Many access switches support port isolation or private VLANs, which stop ports on the same VLAN from reaching each other while still allowing the uplink. If workstations have no reason to talk to each other, an unobservable problem becomes a non-problem. Second, observe it at the access layer. Some access switches can export flows or mirror traffic themselves. At that point their uplinks are no longer the only place you can look. Third, observe it at the host. The built-in firewall on each workstation can log inbound connections. Those logs, collected centrally, show workstation-to-workstation attempts that no network device ever saw. The same limit applies inside a hypervisor, where VMs on one virtual switch talk without touching a physical port. See VM-to-VM visibility. The same limit also applies inside a container cluster, where pods on a node reach each other through the node's own networking, covered in pod-to-pod visibility.
What to log for every zone pair
With the observation points chosen, decide what to keep. For every pair of zones, three kinds of record earn their storage.
- Allowed sessions, with bytes. Source host, destination host, destination port, protocol, bytes in each direction, start time, and duration. Bytes are not optional: the difference between a workstation checking a share and a workstation copying the entire share is invisible without them.
- Denies. Every session the boundary refused, with the same fields plus which rule refused it. A deny on a zone pair marked Block in your matrix is a free alarm. It fires exactly when something tries to do what nothing legitimate should. Keeping the rule identifier in the record lets you tell a noisy old rule from a new event. The rule hit analysis guide covers reading those counts.
- New pairs. The first time a given source host talks to a given destination host across a boundary, and the first time a given port is used between two zones. These are derived from the first two kinds of record, and they are the ones you will actually alert on.
A few practical points. Send everything — firewall session logs, flow records, ACL hits — to one collector. That way, a question about a zone pair has one place to be asked. Keep at least several weeks of history; the value of "have we ever seen this before?" depends entirely on how far back "ever" goes. And do not filter at the source: a record discarded because it looked boring is one you cannot consult during an incident.
Tip: when you enable session logging for allowed traffic on a boundary firewall, watch the firewall's CPU and the log pipeline for a day. Logging every allowed session on a busy boundary is a real load. If it hurts, log the sensitive pairs first (anything into the server and management zones) and add the rest gradually.
Turning addresses into zones
Flow records and firewall logs contain IP addresses. Your matrix is written in zones. Something has to translate, and that something is a lookup table you maintain by hand: every subnet, and the zone it belongs to.
# subnet zone 10.10.0.0/22 users 10.10.4.0/24 users-vpn 10.20.0.0/24 servers 10.20.1.0/24 servers-db 10.99.0.0/24 management 10.30.0.0/24 iot-building 192.168.200.0/24 guest
Most flow collectors let you define named groups of addresses and report on traffic between groups; if yours does, this table becomes those groups. If it does not, a small script that adds a src_zone and dst_zone column to each record before storage does the job. Either way, once every record carries its zones, the matrix stops being a document and becomes a query. Keep the table with your network documentation and update it in the same change that creates or retires a subnet. An unmapped subnet shows up as traffic from "unknown," which is useful only if you notice.
Building a baseline per zone pair
A baseline is a description of what normal looks like, specific enough that you can tell when today is different. For inter-zone traffic, build one per zone pair with four questions. Which destination ports are used? How many distinct source hosts use them? How many distinct destination hosts do they reach? How much data moves per day? Collect for a few weeks — long enough to include a month-end, a patch cycle, and at least one weekend — before you trust it.
Here is an illustrative query in the shape most collectors support, written as SQL for readability. Your collector's language will differ; the structure is what matters.
-- One row per zone pair and destination port, last seven days
SELECT src_zone, dst_zone, dst_port,
COUNT(*) AS sessions,
COUNT(DISTINCT src_ip) AS distinct_sources,
COUNT(DISTINCT dst_ip) AS distinct_destinations,
SUM(bytes) AS total_bytes
FROM flows
WHERE start_time >= NOW() - INTERVAL '7 days'
AND src_zone <> dst_zone
GROUP BY src_zone, dst_zone, dst_port
ORDER BY src_zone, dst_zone, sessions DESC;
Run it and you get the shape of your east-west traffic. For the users-to-servers pair on a typical small network it might read something like this (the figures are illustrative):
| Port (service) | Sessions / week | Distinct sources | Distinct destinations | Notes |
|---|---|---|---|---|
| 443 (web applications) | 84,000 | 212 | 6 | Every workstation, a handful of app servers |
| 445 (file sharing) | 31,000 | 210 | 2 | The two file servers |
| 53 (DNS) | 390,000 | 214 | 2 | Both domain controllers |
| 88, 389 (authentication, directory) | 52,000 | 214 | 2 | Both domain controllers |
Notice what the baseline gives you beyond port numbers. The users-to-servers pair has about two hundred sources and a single-digit number of destinations, every week. That shape — many-to-few — is the fingerprint of a client-server relationship. The management-to-servers pair has the opposite fingerprint: two or three sources reaching dozens of destinations on administrative ports. When a pair's shape changes — a single workstation reaching dozens of destinations, say — you notice it even if every individual port is familiar. Save each pair's baseline as a short table and re-run the query monthly.
Alerting on new pairs and new ports
Volume thresholds are the wrong tool for inter-zone traffic; legitimate volumes swing wildly with backups, updates, and month-end. The right trigger is novelty: something crossing a boundary that has not crossed it before. Three alerts cover most of the ground.
- Any traffic on a Block pair. If the boundary firewall reports a deny, log it and count it. If it reports an allow, or a flow record shows the conversation completing, your segmentation has a leak. In that case, the segmentation verification guide is your next stop.
- A new destination port on an Allow pair. Users reaching servers on a port never seen in the baseline — particularly an administrative port — is the alert most worth tuning carefully.
- A new source host on a narrow pair. The management-to-servers pair has a short list of legitimate sources. A new one is either a new admin machine that belongs in the management zone, or a machine that is not an admin at all.
The query behind "new" compares today against history. Illustratively:
-- Zone-pair + port combinations seen today but not in the prior 30 days
SELECT t.src_zone, t.dst_zone, t.dst_port, t.src_ip, t.dst_ip, t.bytes
FROM flows t
WHERE t.start_time >= CURRENT_DATE
AND t.src_zone <> t.dst_zone
AND NOT EXISTS (
SELECT 1 FROM flows h
WHERE h.src_zone = t.src_zone
AND h.dst_zone = t.dst_zone
AND h.dst_port = t.dst_port
AND h.start_time BETWEEN CURRENT_DATE - INTERVAL '30 days'
AND CURRENT_DATE
);
Two refinements keep this from paging you for nothing. Require a confirmation window. Alert on a new port only once a session on it has completed or carried more than a small number of bytes. That way, a single stray packet does not count. And require persistence for the lower-priority cases: a new source on an Allow pair might wait until it has been seen twice in an hour. Expect the first weeks to be noisy as you discover dependencies nobody documented — a backup agent, a monitoring probe, a scanner in finance. Each one either updates the matrix or opens an investigation; both are the right outcome.
The matrix as a living dashboard
Once records carry zones and the baseline exists, the zone-pair matrix can be drawn on a screen with live data in every cell. Each cell shows the intended state from your matrix (Allow, Log, or Block) alongside what was observed today. It shows sessions, bytes, the number of distinct sources, and two counters that matter more than the rest — new ports and denies. Read it with a simple rule:
| Cell state | What it means | What to do |
|---|---|---|
| Allow, traffic within baseline | The pair is behaving as designed | Nothing; glance at it weekly |
| Allow, new port seen | Something crossed on a port not in the baseline | Identify the source host and the service; update the baseline or investigate |
| Log, traffic present | A tolerated pair is in use | Review monthly; the goal is to move the pair to Allow with ports, or to Block |
| Block, denies counted | The boundary is doing its job; something tried | Note the source; a rising count from one host is worth a look |
| Block, allowed traffic observed | A leak: the rule is missing, misordered, or bypassed | Treat as a segmentation failure today |
| Any pair, source count spike | The many-to-few shape has changed | Find the host behind the change before anything else |
The dashboard earns its keep with a short ritual: fifteen minutes each week looking only at cells that changed. That habit catches drift before drift becomes an incident. And it gives you something to show a manager that needs no explanation: a grid, mostly calm, with the occasional cell asking for attention.
Walkthrough: from raw flows to one dashboard cell
A concrete run-through makes the pieces click. On a Tuesday morning the users-to-servers cell shows one new port. You open the detail:
src_zone dst_zone dst_port src_ip dst_ip bytes first_seen users servers 5985 10.10.1.57 10.20.0.11 48,120 08:42 users servers 5985 10.10.1.57 10.20.0.12 51,004 08:44 users servers 5985 10.10.1.57 10.20.0.14 47,880 08:47
You see one source host in the user zone, three destinations in the server zone, an administrative port (remote management over HTTP), a few minutes apart. The port has never appeared on this pair in thirty days of history. The first check is the easiest one: who is 10.10.1.57? The inventory says it is a laptop issued last week to a new member of the IT team. They have been doing exactly what the traffic suggests — managing servers — from the user VLAN, because nobody told them about the jump host.
A good outcome, and it still demands action. The laptop moves to the management zone (or the admin learns to use the jump host). The users-to-servers rule for that port stays Block. And the baseline is not updated to include it, because the traffic was never supposed to be there. Had the inventory instead said "receptionist's desktop," the same three lines would have been the opening of an incident. In that case, the lateral movement signals guide describes what to look for next.
Where to go next
You now know where inter-zone traffic can be seen — the routing point, a boundary firewall, flow export, and mirror ports. You also know where it cannot: inside a single VLAN, a virtual switch, or a container node. You know the three records worth keeping per zone pair. You know how to build a baseline whose shape tells you as much as its ports, and how to alert on novelty rather than volume. The matrix you drew as a plan is now a dashboard that reports.
The next question is whether the boundaries you are watching actually hold. Verifying Segmentation Actually Works shows how to test that from your own network. The rest of the topic lives on the East-West Visibility & Lateral Movement pillar page.
Put this into practice with SysPeep Monitor
Every observation point in this guide sits on a device you should already be watching. SysPeep Monitor polls your core switch and boundary firewalls over SNMP and graphs each VLAN interface at the zone edges. It alerts when a boundary link drops or its traffic changes shape. That is the health layer underneath any flow-based view. The free trial installs on any Windows machine.
Download Free TrialLearn About SysPeep