Home › Topics › Monitoring Your ISP & Holding Them Accountable › Two Providers, One Dashboard

Two Providers, One Dashboard: Verifying Real Redundancy


The pitch for a second internet provider writes itself: if one goes down, the other carries you. So you sign with a second company, plug a second circuit into the router, watch traffic fail over in a test, and file the subject under "solved". Then one wet Tuesday a contractor's excavator finds a fiber bundle three streets away. Both circuits, from two different companies on two different invoices, die in the same second. I have watched that happen. The redundancy turned out to be two logos sharing one trench.


Redundancy is a claim, and claims get verified. This guide treats a two-provider setup the way the rest of this pillar treats a single circuit: as something to measure rather than trust. You will learn to run identical measurements on both circuits so the providers can be compared honestly. You will learn to build a per-provider scorecard that settles arguments and to ask whether your "diverse" paths are actually diverse. You will learn to watch the standby and the failover as first-class citizens. And, because honesty cuts both ways, you will learn to recognize the cases where a second full provider is not worth its cost.


In this guide

  • One measurement program, two circuits
  • The per-provider scorecard
  • Keeping the comparison fair
  • The diverse-path illusion
  • What you can verify yourself
  • The standby is a real service: watch it like one
  • Redundancy degraded: a first-class event
  • The quarterly redundancy review
  • When the second provider is not worth it

One measurement program, two circuits


Everything starts from a rule so simple it is routinely skipped: measure both circuits identically. Same probe types, same cadence, same external anchors, same retention, same rollups. The moment the measurements differ — circuit A probed every minute toward two anchors, circuit B pinged occasionally toward whatever was handy — every comparison you make afterward is guesswork. Identical instrumentation is what lets you say "circuit B loses packets at triple the rate of circuit A toward the same targets over the same weeks" and have the sentence mean something.


The full measurement kit includes per-minute availability probes, latency and loss toward each provider's gateway and toward shared external anchors, and scheduled throughput samples. It is built in measuring your circuit yourself; run it twice, once per circuit. The one genuinely fiddly part is making sure each circuit's probes actually leave through that circuit. Failover routing would otherwise send every probe down whichever line is primary and leave the standby's graphs describing nothing. The guide to multi-WAN visibility covers those mechanics, and this guide assumes them. One nuance deserves a flag: the two circuits may be different technologies with different natural personalities. A dedicated fiber line and a shared cable service will not produce twin graphs on their best day. Identical measurement does not mean expecting identical results. It means the differences you see are real. That is exactly what you need when deciding which provider deserves the primary role, the renewal, or the boot.


The per-provider scorecard


With identical measurements flowing, comparison becomes a monthly five-minute ritual instead of a debate about impressions. Keep a small scorecard per provider — one table, updated when you do the monthly rollup. It holds the numbers that describe both the circuit and the company behind it:


MeasureCircuit ACircuit B
Availability, this quarter (your probes)99.98%99.71%
Median / p95 latency to anchors12 / 19 ms24 / 61 ms
Loss to anchors, worst week0.1%1.4%
Throughput samples vs. ordered rateconsistently nearfades in evenings
Incidents filed this quarter16
Median time to restore (your timestamps)40 min4.5 hours
Unannounced maintenance events02
Credits claimed / honored0 / 02 / 1

The top half comes straight from monitoring. The bottom half comes from each circuit's evidence file, which you keep per circuit for exactly this reason. Notice what the scorecard measures that graphs alone cannot: the company's behavior. Time-to-restore, honesty about maintenance, whether credits get honored without a fight — these are provider qualities, not circuit qualities. They are what you are really shopping between at renewal. A scorecard like the example above answers questions that otherwise dissolve into anecdote. Which circuit should be primary (A)? Which provider gets the harder renewal conversation (B)? And is "they're both fine" true (no)?


Keeping the comparison fair


A scorecard invites a subtle unfairness that will skew every conclusion if you let it: in a primary-standby design, the two circuits are not living the same life. The primary hauls the whole office's traffic all day; the standby carries nothing but probe packets. And an unloaded circuit looks better than it is: latency graphs are flat and loss sits at zero partly because nothing is stressing it. Congestion-shaped problems — the evening slowdowns, the buffering under load that users would actually suffer — simply cannot appear on a line carrying no load. Promote that "perfect" standby during a real outage and you may discover its personality the hard way. (There will be an audience.)


Three habits restore fairness. First, lean on the scheduled throughput samples: they briefly load the standby on purpose, so at least one measurement exercises it beyond pings. Second, consider an occasional planned role swap — run production over the standby for a business day during a low-stakes window. One day of real load tells you more about that circuit than a quarter of idle graphs. And the swap doubles as a live failover rehearsal. Third, compare like windows with like: the same weeks against the same anchors for both circuits, steady state against steady state. Do not compare circuit A's quiet month against the week circuit B was installed, relocated, or under repair. None of this rescues a truly outclassed circuit, but it keeps you from promoting a circuit whose chief virtue is never having been tested.


The diverse-path illusion


Now the uncomfortable part. Two providers on two invoices can still share almost everything that matters physically. The last few hundred meters are the classic trap. Both services may enter your building through the same conduit, riser, and wall. In that case, one backhoe, one flooded vault, one unlucky delivery truck versus a pole, and both die together. One step up, both may lease capacity on the same regional fiber ring or the same wholesale "last mile" operator. Reselling each other's infrastructure under different brands is normal industry practice, not scandal. With that shared infrastructure, a single cut upstream takes out both services at once. Diversity is a property of paths, not of invoices.


Northgate Retail ran its head office on two circuits from two providers and had a framed failover test result to prove it. Over one winter the evidence files for both circuits recorded three outages in the same ten-minute windows, each closed by each provider as "no trouble found". A traceroute out of each circuit met at the same address by hop three. A written diversity request produced, after some weeks, the information that the second provider leased its last mile from the first. The sales conversation had described that as "fully independent". The failover had worked flawlessly all three times. It had simply had nowhere to go.


You cannot see most of this from your desk, so part of the verification is asking. Put the questions to each provider in writing, before signing where possible: Where does the service physically enter the building, and can it enter somewhere else? Whose infrastructure carries the last mile — yours or a wholesale partner's? Where is the serving node or exchange, and is it the same one the other service uses? Can you provide a diversity statement relative to this other named circuit? Expect imperfect answers — sales layers do not always know, and some detail is legitimately confidential — but the shape of the response is data. A provider who can talk fluently about entry points and diversity is offering a different product from one who answers every question with "it's redundant, don't worry." If real path diversity is a hard requirement, say so in the contract conversation and get what was promised in writing. A promise that lives only in a phone call cannot be produced later.


What you can verify yourself


Asking is half; testing is the better half. Three checks cost you an afternoon and no goodwill at all.


Walk the building

Physically follow each service from its demarc backward: which wall does it pierce, which riser does it ride, where does it leave the property? You are looking for the single point — the one conduit, the one pole, the one basement corner below the waterline — where both services keep company. Ten minutes with a flashlight settles what no brochure will. Two entrances on opposite walls is real diversity you can photograph. Two cables snuggling through one duct is a fact worth knowing before you rely on it.


Trace the paths out of each circuit

A traceroute (on Windows, tracert) lists the routers a packet crosses on its way to a destination. Run it toward the same anchor twice — once forced out each circuit — and compare the early hops:


C:\> tracert -d 198.51.100.53        (via circuit A)
  1     1 ms   203.0.113.1      <- provider A gateway
  2     4 ms   203.0.113.65     <- provider A metro
  3     9 ms   198.51.100.7     <- regional carrier
  ...

C:\> tracert -d 198.51.100.53        (via circuit B)
  1     2 ms   192.0.2.1        <- provider B gateway
  2     6 ms   192.0.2.33       <- provider B metro
  3    11 ms   198.51.100.7     <- the same regional carrier

Read the annotation: the first hops differ — each circuit rides its own provider's edge, as advertised. But by hop three both paths land on the same address of the same regional network. Everything from there outward is shared, and a bad day on that carrier is a bad day on "both" of your providers at once. This does not make the second circuit worthless — it still protects against either provider's own failures. But the trace draws the true boundary of your redundancy, which is exactly what you wanted to know. Capture these traces when everything is healthy and file them as part of your path baselines. The guide to building path baselines shows the discipline. That way, "the paths converged at hop 3" is a recorded fact, not a memory.


Interrogate your own history

The correlation test runs itself once you keep evidence files: scan both circuits' incident logs for failures in the same window. Independent circuits should almost never fail together; even one simultaneous unexplained outage is a loud hint of shared infrastructure, and two is a finding. This is the cheapest diversity audit there is, because the data accumulates on its own. It is the one test that catches sharing nobody would ever have told you about.


Tip: label everything by provider and circuit ID, not by "primary" and "backup." Roles swap, sometimes automatically. A dashboard that says "backup down" when the traffic is currently riding that circuit will mislead exactly the person answering the page at 3 AM. Identity is stable; role is weather.


The standby is a real service: watch it like one


In many two-provider setups, one circuit carries the traffic and the other waits. The waiting one develops a peculiar failure mode. It can be dead for weeks and nobody notices, because nothing depends on it. Then the primary fails, the failover fires into a dead line, and you experience both outages at once. So the standby gets the full measurement program, forever. Its probes run every minute exactly like the primary's. Its alerts page with real urgency. And its failures get filed in its own evidence file with tickets opened just as promptly. Failover behavior itself — drills, detection tuning, per-circuit dashboards — is multi-WAN visibility's territory. One sentence covers this guide's stake in it: a failover you have never watched succeed under monitoring is a hope, not a design.


The accountability angle is the part teams miss. You are a paying customer of the standby's provider, with the same SLA a primary customer has, and "we weren't using it that week" does not discount the invoice. Standby outages count against their availability target, qualify for their credits, and belong on their scorecard. Providers do not know or care which of their circuits you route traffic over — and neither should your measurement program.


Redundancy degraded: a first-class event


When one circuit of two fails, users feel nothing — failover did its job — and that painless success is a trap. Your risk state has changed completely: you are one failure from dark, running unprotected for however long repair takes. Treat redundancy degraded as its own alert class, urgent even at 3 AM precisely because no user will ever report it. The event deserves the same response choreography as an outage. Open the ticket with the failed circuit's provider immediately (their repair clock starts then, not when you get around to it). Log the entry in that circuit's evidence file. Note how long you ran unprotected — exposure hours are a number renewal meetings respond to. The alert-design mechanics live with the multi-WAN guide; the accountability rule lives here. A standby failure is a real incident with a real provider, prosecuted with the same evidence discipline as any outage that made users shout. Handled this way, even your redundancy failures generate leverage.


The quarterly redundancy review


Diversity is not a fact you establish once; it erodes. Providers re-route, resell, and consolidate infrastructure without sending you a card. Construction rearranges the streets your fibers cross. And a path that was diverse at signing can quietly converge a year later. So the verification becomes a small recurring ritual — an hour a quarter:


  • 1. Update both scorecards from the monthly rollups and the evidence files, and note which provider drifted which way.
  • 2. Run the correlation scan: any incidents on both circuits in the same window this quarter? Investigate every match.
  • 3. Re-run the traceroutes out of each circuit toward your anchors and compare against the filed baselines — paths that used to diverge and now converge are news.
  • 4. After any building or street construction, re-walk the physical entries.
  • 5. Count the quarter's failover events and unprotected hours, and confirm the standby carried real load at least once — via throughput samples or a role-swap day.
  • 6. Decide, on the evidence: does the primary keep its job, and does the second circuit still earn its invoice?

Item six is the point of the whole exercise. Redundancy is a purchase that must keep re-justifying itself, and the review is where it does, or does not.


When the second provider is not worth it


Now the honest part. A second full provider is a significant recurring cost, and there are situations where it buys far less than the invoice implies. The first is the one this guide taught you to detect: unfixable shared risk. Both services may only be able to enter your building through one conduit and reach the world over one regional network. That is a common reality outside dense urban areas. In that case, the second provider protects against provider-side failures only. Your dominant risk (the physical path) remains single. Paying full price for half the protection may still be rational; doing it unknowingly is not.


The second is when a cheaper tool covers your actual need. A wireless backup — an LTE or similar cellular link — is physically diverse by nature (no trench, no conduit, different failure weather entirely). It costs a fraction of a second wired circuit. Its tradeoffs are real: less throughput, variable latency, often a usage cap that rules out running the whole office on it for a week. But if the honest requirement is "email, card terminals, and remote access must survive a backhoe," a well-monitored cellular backup meets it. Decide by requirement, not by symmetry. List what genuinely must survive a failure and for how long, then buy the cheapest arrangement that covers the list. Sometimes that is two full providers with verified diverse entries; often it is one good provider — chosen with scorecard evidence — plus wireless backup and honest expectations. The redundant answer is not automatically the responsible one; the measured answer is. And whichever you choose, it goes on the dashboard: an unmonitored backup of any kind is a rumor with a monthly fee.


Where to go next


Redundancy you have verified, with identical measurements, a scorecard, traced paths, and a watched standby, is a different asset from redundancy you were sold. It fails less mysteriously, costs what it should, and gives you two providers who each know you keep score. It is also, as far as you can prove, two paths and not one trench. The same program scales down to one circuit and up to many, and the rest of it lives in Monitoring Your ISP & Holding Them Accountable: independent measurement, SLA literacy, evidence files, and escalation that spends them.


Put this into practice with SysPeep Monitor

Two circuits only count as redundancy if both are watched all the time. SysPeep Monitor probes each circuit separately from one Windows console. It offers per-minute checks, per-circuit dashboards, and alert rules that treat a dead standby as the emergency it is. Its reports are ready for the next provider conversation. The free trial sets up in minutes.

Download Free TrialLearn About SysPeep