East-West Traffic: Watching What Never Crosses the Firewall
At a quarter past nine the file server goes to syrup. Saves hang, the shared drive takes a full minute to list a folder, and the tickets arrive in a cluster. You check your graphs. Internet link healthy, firewall CPU idle, every server answering pings. The slowness is real; you can feel it from your own desk. It is just happening somewhere none of your graphs look: between two switches, on traffic that never leaves the building.
That traffic is called east-west, and on most networks it is both the majority of the bytes and the least observed. I spent my first few years looking at the firewall for everything, because the firewall was where the graphs were. This guide explains what east-west traffic is and why the firewall, the place everyone looks first, is structurally unable to see it. It covers five practical ways to watch it, ranked from free-and-coarse to genuine investment. By the end you will have a starter plan that fits a small shop and a clear idea of what "suspicious" looks like once you can finally see.
In this guide
- North-south, east-west, and why the names matter
- Why the firewall can't help you here
- Option 1: uplink counters — free and coarse
- Option 2: flow export — the sweet spot
- Option 3: mirror ports — the microscope
- Option 4: ask the endpoints
- Option 5: segmentation as a visibility strategy
- What to actually watch for
- A starter plan for a small shop
North-south, east-west, and why the names matter
Network diagrams traditionally draw the internet at the top and the inside of your network below it. Traffic that moves up and down the page — between your network and the outside world, crossing the firewall at the edge — is called north-south. Traffic that moves sideways, from one internal device to another without ever leaving, is called east-west. The compass is silly, but the distinction it names is the single most useful one in visibility planning. That is because the two kinds of traffic are watched by completely different things.
Almost everything your users consider "the network working" is east-west. So is a good deal they never think about; printers, for instance, gossip constantly, mostly about themselves. The main kinds:
- PC to file server: every open, every save, every roaming profile at login time.
- Server to server: backups, replication, the application server talking to its database — the heaviest flows most networks carry.
- Printer chatter: discovery announcements, status queries, driver downloads.
- The dark one — lateral movement: an attacker who lands on one machine and spreads to others. That spreading is deliberately east-west, because the edge is guarded and the interior usually is not.
Now weigh the two. On many small networks, the big-byte events — backups, machine imaging, file copies, database traffic — never leave the building. Meanwhile, the internet link carries a comparatively modest stream of web and cloud traffic. Yet the internet link is graphed, logged, and alerted on, and the interior is not. You are watching the front door of a building where most of the action happens in the corridors.
Why the firewall can't help you here
To see why east-west traffic is invisible, you need one paragraph on how a switch works. Every network interface has a MAC address — a hardware address burned into the card, like a serial number for the port. A switch learns which MAC address lives on which of its ports. It then delivers each frame (the unit of data a switch forwards) directly: in the sender's port, out the recipient's port, done. No detour, no announcement, no permission asked. When two devices sit on the same switch and the same VLAN, their conversation touches exactly one device: that switch. A VLAN is a virtual LAN — a logical partition of the network that behaves like its own switch.
Your firewall, meanwhile, is a border checkpoint. It inspects and logs what crosses it — typically traffic between inside and outside, or between zones it explicitly separates. Traffic that never approaches the border generates no log entry, no counter, no inspection, nothing. Even when two devices sit on different switches, the frames cross the uplink between switches and still touch nothing but switches. This is not an oversight; it is the design. Switching stays fast precisely because it does not ask a central checkpoint for permission on every frame.
The most sensitive conversations on your network — users to the file server, application to database, server to backup target — are usually the least observed ones. (The firewall log, meanwhile, is immaculate.) In the survey of the eight blind spots, this one is listed first for a reason. The rest of this guide is the fix, in five options ranked by effort.
Option 1: uplink counters — free and coarse
Every managed switch counts the bytes entering and leaving each of its ports, and will report those counters over SNMP. That is Simple Network Management Protocol — the standard question-and-answer protocol network devices speak. If you have any monitoring tool at all, you can graph these today at zero cost. The trick is choosing which ports: graph the uplinks — the ports connecting switch to switch — and the ports feeding your important servers. Those few ports are the bridges between areas of your network. And their counters are a running record of how much traffic moves between those areas.
If you want to know what your tool is reading under the hood, it is two counters from the standard interface table:
# 64-bit traffic counters, one pair per port (IF-MIB) ifHCInOctets 1.3.6.1.2.1.31.1.1.1.6 ifHCOutOctets 1.3.6.1.2.1.31.1.1.1.10
Your monitoring system polls these every few minutes and turns the differences into a rate graph. That is the whole magic.
Tip: make sure your tool uses the 64-bit "HC" (high capacity) counters shown above rather than the older 32-bit ones. A 32-bit byte counter rolls over at 4 GB. On a busy gigabit link that is well under a minute of traffic. The rollover produces graphs with absurd spikes that will send you chasing ghosts.
Be honest with yourself about what this option gives you. It is a water meter on the pipe between two buildings: you learn how much flowed and when, never who was talking or why. But "how much and when" already catches real problems. Examples include the backup that drifted into business hours and the uplink quietly saturated every afternoon. Another is the burst of traffic at two in the morning on a switch that serves nothing nocturnal. Volume and timing are the skeleton of east-west visibility, and they are free.
Bellweather Logistics ran two hypervisors (servers that host virtual machines) in the same rack, each with one network port, both into the same switch. A nightly job copied virtual machine images from one to the other. When its start time was moved earlier it began finishing at twenty to nine instead of six. Every morning the warehouse application, which lived on the receiving hypervisor, crawled for forty minutes. Every morning the internet graph, the firewall graph, and the server pings were untroubled. The two server ports were not uplinks, so nobody had graphed them, and the transfer never crossed anything that was. Someone eventually added both ports to monitoring and watched them sit at line rate from eight o'clock until the job ended. Line rate is the most the port can carry. The fix was a change to a schedule. The diagnosis took three weeks.
Option 2: flow export — the sweet spot
The next step up answers the question counters cannot: who was talking. A flow is a summary record of one conversation. It records which device spoke to which, on which ports, how many bytes and packets, starting and ending when. Many routers and better switches can do this bookkeeping themselves and export the records to a collector program. The collector stores them and lets you search and rank. If interface counters are the water meter, flow export is the itemized phone bill: not a word of what was said, but every call, every caller, every duration.
For east-west visibility, the place to enable it is your core or distribution switch — the crossroads that inter-switch traffic actually passes through. With flows arriving from the core you can see your top talkers, the conversation pairs behind an uplink spike, and brand-new pairs that never talked before. The half-past-nine mystery from the introduction dissolves in about thirty seconds. Sort flows by bytes, and the server-to-server replication is sitting at the top of the list wearing a guilty expression.
Three honest caveats. First, not every switch can export flows — it is a feature of the better ones, so check your model's documentation before planning around it. Second, some hardware exports sampled flows, recording one packet in every N. That is fine for spotting patterns and top talkers, but it is not a complete ledger. Third, you need somewhere for the records to land: a collector, which means a bit of setup and disk. All three caveats are manageable, which is why this option is the sweet spot. You get one configuration change, always-on history, and answers to "who," without drowning you in raw packets. The Flow Monitoring guides cover choosing a flavor, setting up a collector, and reading the results.
Option 3: mirror ports — the microscope
Sometimes a summary is not enough and you need the actual packets. Managed switches can do port mirroring (often called SPAN, from a vendor name for the feature). The switch copies every frame passing through one port onto another port, where a machine running capture software listens to the copy. Mirror the file server's port and you see every conversation that server is having, in full detail, down to the individual request. It is considerably more than you wanted to know about most servers.
This is the microscope, and the right way to use it is surgically. When you have a specific suspicion, mirror a specific port for a bounded time, capture, analyze, and turn it off. It is the wrong tool for standing surveillance — the storage fills fast, and most of what you capture is encrypted payload anyway. And a mirror copy is best-effort: under heavy load the switch will quietly drop copied frames rather than slow real traffic. So the copy is not gospel. Reading captures is its own craft, covered in Packet Capture & Analysis.
One preparation makes this option ten times more useful: write the runbook before you need it. Which switch commands set up the mirror, which laptop has the capture software, which spare port it plugs into, who approves it. During an outage is a terrible time to learn mirror syntax; with a runbook, "let's look at the actual packets" becomes a ten-minute move instead of an afternoon.
Option 4: ask the endpoints
The vantage point almost everyone forgets is the two computers in the conversation, both of which know about it. Every operating system will list its own live network connections. That makes every machine on your network a free east-west sensor — one you interrogate by hand, but free. On a Windows machine:
# Windows: list live connections (run elevated for -b, the owning program) netstat -nob # Linux: same idea, with the owning process ss -tunp
A typical line of output, and how to read it:
TCP 10.0.20.31:52144 10.0.10.5:445 ESTABLISHED [System]
Local machine 10.0.20.31 has a connection to 10.0.10.5 on port 445 — the Windows file-sharing port. So this PC currently has a file server session open. Ten lines like that and you have a small map of who this machine talks to, drawn from the machine itself.
Strengths: free, available everywhere, and uniquely detailed — the endpoint can tell you which program owns each connection, which no switch-side view can. It is the perfect answer to "what is this server talking to right now?" The limits are that it is one machine at a time, a snapshot rather than history, and it depends on the machine telling the truth. A seriously compromised host can lie about its own connections. That is precisely why security investigations prefer network-side vantage points like flows and mirrors. Use endpoint views for operational questions, and treat them as corroborating rather than conclusive evidence for security ones. A witness, not a camera.
Option 5: segmentation as a visibility strategy
The first four options add sensors to the paths traffic already takes. The fifth is sneakier: change the paths. Suppose you place a routed or firewalled boundary between two zones of your network — say, user devices in one VLAN and servers in another. Then every conversation between those zones must cross the device at the boundary. And a boundary device is exactly the kind of thing you can watch: it counts, it logs, it can filter. Traffic that used to slide invisibly from switch port to switch port now files past a checkpoint you own.
This is visibility as an architecture choice rather than a sensor choice, and it is worth taking seriously the next time you redesign anything. It also pays double: the same boundary that gives you visibility can enforce rules. Workstations may reach the file server on file-sharing ports and nothing else, for example. Enforcing those rules converts your east-west blind spot into your east-west security policy. Lateral movement, the attacker technique from earlier, suddenly has to cross a logged, filtered checkpoint instead of an open corridor.
Segmentation is real work, and it is only fair to say so. Someone has to design the zones, and the boundary device has to be sized for the traffic that will now cross it. The migration has a way of discovering undocumented dependencies at inconvenient moments. A thirty-person office does not need a micro-segmented fortress; one users/servers boundary buys most of the visibility value for a fraction of the effort. Treat it as the long game. Adopt it at the next natural moment — a firewall replacement, an office move, a switch refresh — rather than as a project justified by visibility alone.
What to actually watch for
Visibility is only useful if you know what "wrong" looks like. Once you can see east-west traffic — even just at the counter level — three patterns deserve standing attention:
| Pattern | What it looks like | What it usually means |
|---|---|---|
| Backup flood at the wrong hour | A fat, flat-topped transfer between servers during business hours | A job rescheduled, overrunning its window, or stuck retrying in a loop |
| Workstation-to-workstation chatter | PCs holding conversations with each other instead of with servers | Often someone sharing a folder from their desktop; occasionally worm-style spreading. Either way, look |
| New talkers to sensitive servers | A device that never touched the finance server opening connections to it | An integration nobody announced — or reconnaissance. Both deserve a question |
The middle row rests on one rule: on a well-behaved business network, workstations talk to servers, and rarely to each other. Client-to-client conversations do have innocent explanations — peer-to-peer features, some conferencing tools, the occasional desktop-hosted share. But client-to-client conversations are uncommon enough that peer chatter is a smell. It is one of the few patterns where a junior admin with a flow report can spot the same thing a security product would flag.
All three patterns share a prerequisite: a baseline, a record of what the traffic looks like when nothing is wrong. Spend a week just looking before you judge anything. A whole week. Not a busy afternoon and a guess. Learn the daily shape of your uplinks — the morning logon bump, the backup plateau at night, the lunchtime dip. Anomalies are only visible against a known normal, and the week you spend learning normal is the highest-value week in this whole subject. It also pays off in reverse. Once you know the shape, a missing feature — the backup plateau that failed to appear last night — is just as loud a signal as any spike.
A starter plan for a small shop
The pragmatic sequence, sized for one admin with a normal workload:
- Step 1 — today, free: list your inter-switch uplinks and key server ports, and add their traffic and error counters to monitoring. This is an hour of work.
- Step 2 — this week, free: baseline. Watch the daily shape, note the backup windows, and write down what normal looks like while it is fresh.
- Step 3 — this month, modest: check whether your core switch supports flow export. If yes, stand up a collector and enable export on the core alone — one device covers most east-west crossroads.
- Step 4 — before you need it, free: write the mirror-port runbook: commands, capture laptop, designated port, approval. File it where you will look for it during an outage.
- Step 5 — the long game: put a users/servers boundary on the roadmap for the next natural redesign, and let visibility be one of its selling points.
Notice what this plan does not include: buying anything. Every rung except the last is configuration and habit, which is the general theme of blind-spot work. The visibility you lack is mostly visibility you have not switched on. Step 4 is the one people skip, because its only customer is future-you, mid-outage, who is not in the room to argue for it.
Where to go next
East-west traffic is the first and usually the largest of the network blind spots, but it has siblings: encrypted payloads, unmanaged devices, isolated segments. The survey in The Eight Places Your Monitoring Can't See helps you rank which ones matter for your network. Then What Encrypted Traffic Still Tells You picks up the natural next question. Fine, I can see the conversations, but what can I learn when they are all encrypted? The rest of the Network Blind Spots guides take it from there. The next time the file server goes to syrup at a quarter past nine, you will have a graph that is as unhappy as you are. And that is the entire point.
Put this into practice with SysPeep Monitor
The first rung of east-west visibility is plain interface monitoring, and that is bread-and-butter work for SysPeep Monitor. It auto-discovers your switches and their ports, graphs every uplink over SNMP, and alerts when traffic or errors break pattern. So the quarter-past-nine flood shows up on a chart before it shows up in tickets. The free trial installs on any Windows machine.
Download Free TrialLearn About SysPeep