Monitoring WAN Links to Remote Sites (Without Staff There)
Somewhere on your network map is a branch office you have visited twice. You went once to install everything, once to fix the thing that broke the week after. There is no IT staff there. The closest thing to remote hands is an office manager who can, with coaching, tell you whether "the light is blinking normal blinking or angry blinking." And this morning the whole site just went gray on your dashboard. That raises a question that will define your next four hours: gray how?
Because from headquarters, every branch failure looks exactly the same. Circuit cut by a backhoe? Gray. Router hung? Gray. Power out in the whole building? Gray. Someone unplugged the rack to run a floor polisher? Also gray. The failures look identical but their fixes could not be more different. One is a provider call, one is a reboot, one is a call to the landlord, and one is a laminated sign. Guess wrong and you burn an afternoon driving to a site whose actual problem was a provider outage you could have reported from your chair.
This guide is about making that gray tile speak. You will learn where to put your measurement points so a branch is visible from more than one angle. You will learn how to tell site-down from circuit-down from power-down using evidence you can collect in advance. You will learn how to stop one dead circuit from paging you forty times. And you will learn what belongs in the standard monitoring kit for every unstaffed site you are responsible for.
In this guide
- The branch problem: no hands, no eyes
- Probe placement: the view from HQ vs. the view from inside
- Site-down, circuit-down, or power-down?
- One circuit, one alert: dependency rules
- The backup path: cellular and out-of-band visibility
- The remote-site monitoring kit
- Where to go next
The branch problem: no hands, no eyes
At headquarters, monitoring is generously supplemented by humans. Something misbehaves and someone smells the smoke, hears the fan screaming, or notices the switch with no lights. Tickets arrive with useful clues attached. When in doubt, you walk over.
An unstaffed site strips all of that away. Monitoring is not a supplement there; it is the only witness. Whatever evidence you will have during an incident is exactly the evidence you arranged to collect in advance — nothing more arrives later except by car. That inverts the usual economics of monitoring. At HQ, an extra check is a convenience. At a branch, an extra independent check is the difference between a diagnosis and a road trip.
The key word is independent. Ten checks that all cross the same WAN circuit are one check wearing ten hats. The circuit dies and all ten go gray together, telling you nothing you did not already know. Useful branch monitoring is built from evidence sources that fail separately — different vantage points, different dependencies, different power sources. The rest of this guide is really just applications of that one idea.
Probe placement: the view from HQ vs. the view from inside
There are two places to watch a branch from, and the golden rule is that they are complements, not alternatives.
The view from HQ is your central monitoring system reaching across the WAN. It pings the branch router's WAN and LAN addresses, polls its interfaces over SNMP, and probes a device or two beyond it. This view is cheap — no branch hardware needed. It measures the thing users at other sites experience: whether the branch is reachable through its circuit. Its blind spots are structural. Every measurement includes the circuit, so a slow circuit poisons every number. The branch server looks slow even though it is fine, because your probe rode a congested line to reach it. And the moment the circuit dies, this entire view goes dark at once; it can tell you the branch is unreachable, but never why.
The view from inside comes from a small measurement presence at the branch itself. It is a lightweight probe or agent on some always-on box, reporting back to your central system. (Many monitoring platforms are built around exactly this pattern: one central service, small remote probes at each site.) The inside view keeps watching the branch LAN when the circuit is down. It catches purely local sickness — the dying switch, the full server disk, the access point that reboots hourly. It can also test outward, measuring the internet as branch users actually experience it. Its blind spot mirrors the first. When the WAN is down it may be unable to report home in real time. So its live value during a hard outage is limited. A good probe, though, buffers its measurements and backfills the history when the line returns. That turns even an outage into a documented outage.
| Question | View from HQ | View from inside |
|---|---|---|
| Is the branch reachable through its circuit? | Yes — this is its home turf | No — wrong side of the question |
| Is the circuit slow or lossy right now? | Yes, via probes through it | Yes, testing outward |
| Is a branch-local device sick? | Only while the circuit is up, and colored by it | Yes, always |
| What is happening during a WAN outage? | Nothing — all gray | Keeps recording; reports when line returns |
| What does the internet feel like from the branch? | No — wrong vantage point | Yes |
| Hardware cost at the site | None | One small always-on box or agent |
There is one more trick hiding in the inside view: its silence is a measurement. A site probe that reports every minute is also a heartbeat. When HQ stops hearing it, something between here and there is wrong, even before any check formally fails. You will use that heartbeat as a primary evidence source in the next section.
What to run the probe on, and who polls what
The probe needs nothing glamorous: the branch server if one exists, a small fanless box, or a lightweight virtual machine on anything always-on. What matters is that it is boring, wired, and plugged into the UPS. Never on Wi-Fi — you do not want your witness testifying about its own connection problems. The UPS lets it survive the opening seconds of a power event long enough to shout. Once the probe exists, divide the labor sensibly. Let the probe do the site's local polling — the switch, the server, the printers — and send HQ the results. Do that rather than having the central system poll forty branch devices individually across the WAN. That division keeps monitoring chatter off a link you may be paying for by the megabit. It keeps local polling running through WAN outages and shrinks every measurement's error bars. Polling a switch from ten feet away is not colored by circuit weather. The central system keeps for itself exactly the checks that only make sense from outside: the probes through the circuit, and the heartbeat listening.
Site-down, circuit-down, or power-down?
Now the main event. The three failures that gray out a branch are distinguished not by any single check but by the pattern across independent evidence sources. Set up the sources in advance and the pattern reads like a verdict.
Here are the evidence sources worth having, roughly in order of value. First is the HQ-side circuit state. On a point-to-point circuit or a site-to-site VPN, the HQ end has an interface or tunnel whose state you can read locally. No branch cooperation is required. Next is the site probe heartbeat, as above. Then comes the UPS at the branch, speaking SNMP either directly or through the box it is attached to. Its "on battery" event is the single most diagnostic message an unstaffed site can send. And there is the branch switch, whose port states report indirectly on everything plugged into it. None of these is exotic; together they see around corners.
Three gray tiles, three different stories
Story one: the site probe heartbeat stops, and HQ-side probes through the circuit all fail. But the last thing in the log before silence is the branch UPS reporting on battery. Verdict: power. The building lost electricity, the UPS gasped its warning, and everything at the site is now either on battery or off. Nobody needs to inspect a router. You call the landlord or the utility, tell the branch manager the outage is bigger than the computers, and watch for the UPS's return.
Story two: heartbeat stops, circuit probes fail, HQ-side tunnel drops — and there was no power warning at all, no on-battery event, nothing. The failure arrived from outside the building. Verdict: circuit (or the provider's side of it). Your move is the provider call, armed with the circuit ID and the timestamped graphs from your baseline register. The evidence habits from WAN circuit health pay off precisely here, at a site you cannot walk to.
Story three: the circuit probes to the branch router fail… but there are signs of life. The site probe's heartbeat is still arriving (over a backup path, or because only some things died). Or the router answers on its LAN address via the tunnel while its WAN side is dead. Or the switch is up but the router's port on it went down. Verdict: something local — the router itself, its power brick, a cable. This is the one that may genuinely need hands. But now you are driving with a diagnosis and the right spare in the trunk instead of a mystery.
| Evidence | Power-down | Circuit-down | Router/site fault |
|---|---|---|---|
| UPS "on battery" just before silence | Yes — the tell | No | No |
| Site probe heartbeat | Stops (or runs on UPS briefly) | Stops (buffers locally) | Often still arriving |
| HQ-side circuit/tunnel state | Down | Down | Down or flapping |
| Branch switch port states (when reachable again) | Everything rebooted at once | All normal throughout | One device's port down |
| Your first call | Landlord / utility | Provider, with circuit ID | Remote hands or a drive |
Two timing refinements keep the verdicts honest. First, demand confirmation before declaring anything. A single missed heartbeat or one failed probe cycle is weather, not news, so page only after several consecutive misses. Branch circuits, especially the budget kind, have moods. A circuit that flaps — up two minutes, down one, all afternoon — deserves its own recognition rather than a hundred up/down pages. The HQ-side tunnel state bouncing is the signature. The right response is one "circuit unstable" alert and a provider call about the line, not a stack of resolved-and-reopened tickets. Second, do forensics after recovery: when the site returns, read the uptime counters. Every device at the site freshly rebooted means the whole building lost power, even if no UPS message escaped in time. Switch and server uptimes sailing through the outage untouched mean the circuit alone died while the site hummed along in the dark, unaware. The gray tile keeps its secrets during the outage — but it confesses on the way back up, if you remember to ask.
Tip: rehearse story one before it happens by asking the person who is at the branch — the office manager — to be part of the system. Put a laminated card by the rack: "if IT calls and says the word power, read them the lights on the black box, top to bottom". That converts a well-meaning bystander into a functioning sensor. It is astonishing how much diagnostic bandwidth one calm phone call gains when the questions were agreed on in advance.
One circuit, one alert: dependency rules
A branch site multiplies alerts in the worst way. Behind that one circuit live the router, a switch or three, a server, a NAS, printers, access points, cameras, and door controllers. That is forty monitored things, all reachable only through one line. The night the circuit dies, naive monitoring pages you forty-one times. It pages once for the circuit and forty times for devices whose only crime was standing behind it. Do that twice and on-call starts sleeping through branch pages. That is how the server-room flood at a branch eventually goes unread in a pile of routine circuit noise.
The fix is to teach your monitoring the shape of the network. The circuit (or the branch router) is the parent, and everything at the site depends on it. When the parent is down, checks on its children are suppressed or their alerts folded into the parent's. So the page reads the way a colleague would say it: "Branch 4 unreachable — circuit down; 40 dependent devices suppressed." One page, full context, correct target for the response. Every serious monitoring system supports some version of this. At unstaffed sites it is not an optimization but a survival requirement. The general craft of modeling these relationships is covered in dependency mapping for alerts.
One refinement earns special mention for branches: make sure suppression happens at the right parent. If the site has a backup path, "circuit A down" should suppress nothing except circuit A's own dependents-of-one. In that case, the site is still up via B, and the correct alert is the redundancy-degraded class from multi-WAN visibility. Only "no path to site at all" should trigger the big fold. Dependencies drawn casually turn into dependencies that hide real outages.
The backup path: cellular and out-of-band visibility
Branches increasingly carry a second, skinnier way out — typically a cellular modem on the router, sometimes a second cheap wired line. Everything about idle backup paths from the rest of this pillar applies here with extra force. Nobody at the site will ever notice the backup light turn red. Probe the backup path continuously even when idle, alert on its failure at degraded severity, and drill the switchover on a schedule. The full discipline is in WAN failover readiness.
Cellular adds two branch-specific wrinkles. First, signal quality is a health metric: most cellular modems expose signal strength and quality readings. A link that has quietly degraded from good to barely-there signal is a backup that will collapse under real load. Trend signal quality like any other counter. Second, metered data changes the monitoring math: a chatty probe schedule that is background noise on fiber is real money on cellular. So probe the backup at a gentler cadence and keep heavy polling off it. Watch monthly volume so a routing mistake that shifts daily traffic onto the modem announces itself before the invoice does.
The same cellular path, or a small dedicated out-of-band device, can also serve as your emergency side door. It is a way to reach the branch router's console when the main circuit is the thing that died. For monitoring purposes the point is simpler — it is one more independent evidence source. During a gray-tile incident, a heartbeat arriving via the side door while the main path is dark is by itself a diagnosis.
The remote-site monitoring kit
All of the above condenses into a standard kit — the checklist to run against every unstaffed site you inherit or build. None of it is expensive; all of it is cheaper than one wrong drive.
- A managed switch with SNMP enabled — the site's nervous system, reporting port states for everything plugged in.
- A UPS that can talk — on SNMP directly or via an attached device, with the on-battery event alerting; it is your power-versus-circuit tiebreaker.
- A site probe or agent on an always-on box, reporting to the central system every minute — inside view plus heartbeat.
- Circuit probes from both directions — HQ probing through the circuit, the site probe testing outward, each with its own baseline.
- Dependency rules configured — circuit as parent, one page per real problem, backup path modeled correctly.
- The backup path monitored — probed while idle, signal quality trended if cellular, volume watched if metered.
- A current baseline register entry — circuit ID, provider contact, demarc location, normal numbers.
- A labeled rack and a photo — every cable tagged, a picture of the healthy rack stored with the register, so phone-guided hands can find things.
- A site contact card — who has keys, who answers the phone, and the laminated what-to-read-me card by the rack.
- A heartbeat alert — explicit alerting on probe silence, so the absence of news is itself news.
Run the list on one branch this week. The first incident afterward will feel different in a way that is hard to oversell. The tile goes gray, and instead of a mystery you get a story with a verdict attached.
Where to go next
One well-instrumented branch is a pattern; twenty of them are a rollout, with template configs, probe fleets, and dashboards per site. The guides in scaling your monitoring cover growing from one site to many without drowning. For the circuits themselves, the metric set and baseline habits live in WAN circuit health. The rest of this pillar — router health, multi-WAN, failover drills — is at Router & WAN Link Monitoring.
Put this into practice with SysPeep Monitor
The inside view of a branch does not require a server there. SysPeep Monitor runs as one central service on a Windows machine with lightweight remote probes for your other sites. Each probe watches its local network, tests the WAN from the branch's side, and reports home. Dependencies keep a dead circuit down to a single page. Try it on your most annoying branch with the free trial.
Download Free TrialLearn About SysPeep