WAN Circuit Health: The Metrics That Describe a Line
Every network you will ever run has a spot where your cables end and somebody else's begin. The WAN circuit is the line you rent from a provider to connect your site to the internet, or to another site. It is the piece of infrastructure you depend on most and control least. You cannot log into it. You cannot reboot it. And you cannot walk over and wiggle its cables, because most of it is under a road somewhere. When it misbehaves, the only tools you have are measurements and a phone number.
That is exactly why circuits deserve the same disciplined monitoring you give a server. Yet most small networks watch their most expensive line with a single ping check, if that. Then one day users say the internet "feels slow." The circuit becomes a suspect, and nobody can confirm or deny anything. Nobody ever wrote down what normal looks like for that line.
This guide treats the circuit as a first-class piece of infrastructure. By the end you will know what a circuit physically is and where your responsibility ends. You will know the five metrics that describe a circuit. You will know why you need to measure both through the line and at its interface. You will know how to build a baseline register for every circuit you rent. And you will know how to read the graph shapes that spell congestion, line errors, and provider trouble.
In this guide
- What a circuit actually is
- The five numbers that describe a line
- Measuring through the circuit vs. at the interface
- The circuit baseline register
- Reading the graphs: four shapes of trouble
- The circuit-health checklist
- Where to go next
What a circuit actually is
In provider language, a circuit is the delivered service between your premises and their network: one line, one contract, one circuit ID printed on the bill. It does not matter whether the underlying technology is fiber, copper, coax, or a dish on the roof. From your side it is "the thing that carries our packets away," and it can be described by the same handful of numbers.
It helps to walk the physical path once, slowly. Your router — which the provider will call the CPE, customer premises equipment — plugs into the handoff: the physical port where the service is delivered. On modern circuits the handoff is usually an Ethernet jack, sometimes on a small provider-owned box. That box is often called an NID or NTU — a network interface device that terminates their line. That box, or that jack, is the demarcation point, universally shortened to demarc. It is the official boundary where the provider's responsibility ends and yours begins. Beyond the demarc, your packets ride the last mile — the physical run from your building to the provider's nearest facility. They land on the provider edge: the first router that belongs to them, where your circuit joins their network.
Think of it like the driveway from your parking lot to the public road. Everything on your side of the property line — your parking lot, your gate, your stretch of pavement — is yours to fix. One step past the line, potholes belong to the county, and all you can do is report them with a precise location. The demarc is that property line, and the circuit ID is the address the county needs before they will even look at your pothole.
This boundary is not legal trivia; it shapes every monitoring decision in this guide. Any measurement that crosses the demarc mixes your equipment and theirs into a single number. The whole game of circuit monitoring is to design measurements that let you tell, from your desk, which side of the line is sick. The fix is a cable swap on one side and a support case on the other.
The five numbers that describe a line
A circuit is healthy when it is up, fast, clean, steady, and not full. Those five plain words map to five measurable metrics, and together they are the complete vital-signs panel for any line you rent.
Availability is the fraction of time the circuit actually passes traffic. Note the phrasing: passes traffic, not "interface shows up." With an Ethernet handoff, your router's port links to the provider's box a few feet away. So the interface can glow a healthy green while the line beyond that box is completely dead. Real availability is proven by packets that make it to the far side and back, which is why probing (next section) is not optional.
Latency is the round-trip travel time of a packet across the circuit, usually in milliseconds. Every circuit has a natural latency floor set by distance and technology. The floor itself is neither good nor bad — what matters is knowing yours and noticing when it moves. The fundamentals of measuring it well are a topic of their own, covered in how to measure latency properly. For circuit health, the number you care about is latency from your router to the provider edge, and how stable it stays under load.
Packet loss is the percentage of packets that enter the circuit and never come out. A healthy line sits at zero or a hair above it. Even small persistent loss punishes real traffic badly, because TCP — the protocol under most applications — interprets loss as congestion and slows down in response. Users experience a lossy line not as an outage but as everything being mysteriously, unevenly slow, which is the most complaint-generating state a network can be in.
Jitter is the variation in latency from packet to packet. A circuit that delivers every packet in 20 ms is steadier than one that alternates between 5 ms and 60 ms, even though the second has a better average. File transfers barely notice jitter. Voice and video calls live and die by it, because they must play audio at a steady rate no matter when packets arrive. If your users are on softphones or video meetings all day, jitter is a headline metric, not a footnote.
Utilization, in and out is how much of the circuit's capacity you are using. It is deliberately two numbers, because a circuit is two one-way pipes. Many services are asymmetric, with less capacity upstream than down. Even on symmetric lines the traffic itself is lopsided: downloads dominate inbound while backups and cloud sync dominate outbound. Networks regularly saturate one direction while the other idles. If you graph only a combined number, the full pipe hides behind the empty one. Always collect and alert on the two directions separately.
Measuring through the circuit vs. at the interface
Here is the core mental model of this whole guide. There are two places to stand when you measure a circuit, and they answer different questions. Standing at the interface means reading the counters on your own router's WAN port. Standing at the probe means measuring through the circuit — sending test packets across it to a target on the far side. Monitor from both places, always, because each one is blind to exactly the things the other sees.
Through the circuit: probe to the far side
A probe is a small, regular test message — typically a ping — sent every minute or faster to a fixed target on the far side of the circuit. For a point-to-point circuit between two of your sites, the natural target is the far router's WAN address. For an internet circuit, use the provider's gateway: the first address past your demarc, which your router learned when the service came up. Probing across the circuit but no further keeps the measurement honest — it isolates the line itself instead of blending in half the internet. Each probe returns three of your five metrics in one shot. It either answers or it does not (availability and loss). It takes some time to return (latency). And the wobble across many probes is jitter.
C:\> ping -n 20 203.0.113.1
Reply from 203.0.113.1: bytes=32 time=18ms TTL=64
Reply from 203.0.113.1: bytes=32 time=17ms TTL=64
Request timed out.
Reply from 203.0.113.1: bytes=32 time=44ms TTL=64
...
Packets: Sent = 20, Received = 19, Lost = 1 (5% loss)
Approximate round trip times in milli-seconds:
Minimum = 17ms, Maximum = 44ms, Average = 21ms
That is one manual sample; a monitoring system runs the same test around the clock and keeps the history, which is where the value lives. One ping with a timeout means nothing. A graph showing loss appearing every weekday at lunch means everything.
At the interface: read the counters
Meanwhile, your router's WAN interface keeps meticulous books, readable over SNMP (Simple Network Management Protocol, the standard way monitoring tools ask devices about themselves). Four counter families matter for circuit health. Interface status tells you if the port is up. Traffic octets in and out give your utilization, once converted to bits per second. Input errors count packets that arrived damaged — a physical-layer symptom. Output discards count packets your own router dropped because the circuit was too full to send them — a congestion symptom. Counters see things no probe can: exactly how full each direction is, and whether the local physical layer is clean.
Interface WAN1 (to circuit FIB-88214) ifOperStatus: up Utilization: in 87.2 Mbps / out 12.4 Mbps (100 Mbps service) ifInErrors: 0 (steady) ifOutDiscards: 14,882 (climbing during business hours)
Read that sample like an admin: the inbound direction is running close to the line's capacity. And the router is discarding outbound packets it cannot send. Nothing is "down," yet users on video calls are having a bad day. No probe alone would tell you the cause this precisely. And no counter alone would tell you how the line actually feels to traffic crossing it. Together they interrogate each other, and the combinations point at a culprit:
| Probe says | Counters say | Most likely story |
|---|---|---|
| Clean and fast | Clean, low utilization | Circuit healthy — look elsewhere for the complaint |
| No answers | Interface down | Physical problem at your end, the handoff, or the last mile |
| No answers | Interface up, counters clean | Line dead beyond the handoff — provider's side of the demarc |
| Slow, some loss | Utilization at capacity, output discards climbing | You are filling the circuit — congestion on your side |
| Slow, some loss | Low utilization, no errors | Trouble inside the provider network — time for evidence |
| Steady loss at all hours | Input errors climbing | Damaged cable, connector, or handoff hardware — physical |
Notice how many rows end in a clear "your side" or "their side." That is the demarc boundary doing its job: two vantage points, one on each side of the question, triangulating the fault before you touch anything.
The circuit baseline register
Now for the unglamorous habit that makes all of the above usable at 2 AM. Write down what each circuit is and what normal looks like for it. A baseline register is a small table — a wiki page or a document is fine — with one entry per circuit. It exists because the two moments you need this information are the two worst moments to reconstruct it. You are mid-outage, or mid-phone-call with a provider who will not open a case without a circuit ID.
| Field | Example: HQ internet | Example: branch point-to-point |
|---|---|---|
| Circuit ID (exactly as billed) | FIB-88214-HQ | PTP-4471-BR2 |
| Provider & support number | Metro fiber co., number on file | Regional carrier, number on file |
| Service & handoff | Fiber, Ethernet handoff on NID | Ethernet over copper, wall jack |
| Contracted capacity (down / up) | 500 / 500 Mbps | 50 / 50 Mbps |
| Demarc location | Basement closet, rack 2, panel B | Branch storage room, left wall |
| Probe target | Provider gateway 203.0.113.1 | Far router WAN 198.51.100.2 |
| Normal latency to target | 3–5 ms | 11–14 ms |
| Normal busy-hour utilization | in ~40%, out ~10% | in ~25%, out ~20% |
| Notes | NID needs provider to reset | Shares conduit with power — noisy |
The last three rows are the baseline proper: normal latency, normal loss (implied zero unless noted), and normal utilization by time of day. You do not have to guess them — run your probes and counters for a couple of weeks and read the numbers off the graphs. From then on, every judgment call becomes a comparison instead of a debate. Latency of 14 ms is meaningless in the abstract; latency of 14 ms on a circuit whose register says 3–5 ms is a finding.
The register also becomes half of your evidence pack on the day you need to convince a provider that their line, not your network, is dropping packets. The art of making that case persuasively is covered in proving it is the ISP, and it leans entirely on the baselines and graphs you are building here.
Tip: record the circuit ID character-for-character as it appears on the bill. Tape a copy inside the door of the closet where the demarc lives. Provider support systems index everything by that ID, and "the fiber one at our main office" has never once located a circuit. Future-you, standing in that closet with a flashlight, will be grateful.
Reading the graphs: four shapes of trouble
With probes and counters flowing, each circuit produces a small family of graphs. The skill of reading them is mostly pattern recognition, and the good news is that circuits fail in a handful of recognizable shapes. Here are the four you will actually meet.
The flat top: congestion
Utilization climbs through the morning and then stops climbing — not because demand stopped, but because the line is full. The graph plateaus into a distinctive mesa shape at the circuit's capacity. On the same timeline, latency rises and jitter gets ragged as packets queue up waiting for their turn. Loss appears at the worst peaks as queues overflow (those output discards again). The flat top is the signature of a circuit you have outgrown, or of one heavy new flow. That could be a backup job rerouted through the wrong hours, a cloud migration, or someone's video habit. The response is investigation and capacity math, not a support case: this shape is traffic, not fault.
The constant drizzle: line errors
Loss sits at a steady low percentage at every hour — 3 AM and 3 PM alike — while utilization stays modest and latency stays normal. Congestion cannot explain loss on an empty line; physics can. A damaged cable, a corroded connector, water in a junction, a failing optic somewhere along the last mile. Check your side first. If your interface's input errors are climbing too, the damage is close to home — between your router and the handoff. There, you can swap a patch cable in five minutes. Clean local counters plus round-the-clock loss points past the demarc, and your graphs of it are exactly what the provider's technician needs to see.
The step change: the path moved
Latency sits at its baseline for months, then jumps to a new, higher, perfectly stable level overnight. Nothing is broken — everything still works — but the floor moved. This is usually the provider rerouting your traffic: a maintenance, a fiber cut healed the long way around, a change in their network. A step change is worth noticing even when no one complains, because it silently eats the latency budget of everything running over the line. The guide on detecting path changes covers how to catch these shifts systematically and what they mean.
The slow ramp: growth
Busy-hour utilization creeps upward over months — each week's mesa a little taller. Nothing to fix today; everything to plan for. The slow ramp is the friendliest shape on this list because it hands you the one thing IT rarely gets: warning. Circuits take time to order and upgrade. The moment your busy-hour normal crosses roughly the three-quarters mark is the moment to start the paperwork, while the choice of timing is still yours.
The circuit-health checklist
Everything above compresses into a checklist you can run against any circuit — one you already monitor, or one you just inherited. Each row is a question, the metric that answers it, and where the answer comes from.
| Question | Metric | Where it comes from |
|---|---|---|
| Is the local end alive? | Interface status | SNMP on your WAN port |
| Does the line pass traffic? | Probe availability | Ping through to the far side |
| Is it its normal speed? | Latency vs. baseline | Probe history + register |
| Is it clean? | Packet loss | Probe statistics |
| Is it steady enough for calls? | Jitter | Probe statistics |
| Is there headroom, both directions? | Utilization in / out | Interface counters |
| Is the physical layer healthy? | Input errors | Interface counters |
| Are we dropping our own traffic? | Output discards | Interface counters |
| Could I open a support case right now? | Circuit ID + baselines | The register |
If every question has a live answer, the circuit is genuinely monitored. If some are blank, you now know exactly which check to add next. Most of them are a five-minute job in any monitoring tool: one SNMP interface plus one recurring ping.
Where to go next
A circuit never fails in isolation — it terminates on a router at each end. And a struggling router can fake every symptom on this page. The companion guide on router health and the forwarding path covers the box itself. If your site has two circuits for redundancy, the rules change in sneaky ways — multi-WAN visibility shows how to watch two lines as one service. And the rest of the Router & WAN Link Monitoring pillar continues from here into failover testing and remote sites.
Put this into practice with SysPeep Monitor
Every number in this guide already exists in your routers — it just needs collecting. SysPeep Monitor discovers your WAN interfaces automatically. It gathers utilization and error counters over SNMP. It runs continuous ping checks through each circuit for latency and loss. It keeps the history that turns a vague complaint into a one-glance diagnosis. The free trial installs on a Windows machine in minutes.
Download Free TrialLearn About SysPeep