Backup internet and automatic failover for business
Two internet providers can fail together when both circuits ride the same fiber. We build a second path that stays off it: a point-to-point radio link from our own network to your building.
The backup link we built held through a regional fiber cut
The problem
A client that runs around the clock needed internet its critical systems could depend on. It already paid two ISPs, but both of their circuits ran over the same long-haul fiber, so one major outage on that fiber would cut both.
What we built
We put in a high-speed point-to-point wireless link that bypassed the long-haul fiber entirely and connected the site to the nearest of our major points of presence, which terminates at a secure data center. That gave the client a completely independent path, carrying over 2 Gbps in each direction at under 19ms of latency.
What happened
Roughly a year and a half later, a major fiber cut struck the client’s primary circuits and set off an extended regional outage. Our link was untouched, and failover moved the client’s traffic onto it automatically, with no downtime.
The client’s IT Director reported that our link “performed flawlessly during the prolonged outage.”
Five ways a business connection goes down, and what survives each
Each failure removes a different part of the connection, so a second path protects you from the failures it does not share with the first.
Long-haul fiber cut
Takes out: Every circuit riding the cut span, however many companies invoice for them.
Survives it: A path that never touches that span, which means a different physical route.
Carrier-side outage
Takes out: Every customer of one provider at once, when a routing, platform or configuration fault hits that provider’s network.
Survives it: A circuit from a second provider, as long as that provider does not lean on the first one for transport.
Power at the building
Takes out: The edge equipment, and with it every circuit that terminates there.
Survives it: Standby power sized for the edge equipment and tested under load.
Last-mile equipment
Takes out: One path: the radio, optical terminal, handoff port or router interface that failed.
Survives it: The other path, provided it terminates on different equipment.
Building event
Takes out: Everything that enters through the affected space, whether the cause is a contractor in the riser, water in the MPOE or the landlord’s Saturday electrical work.
Survives it: A second path into the building through a different entrance.
Four kinds of diversity, and the one that survives a fiber cut
Two circuits from two companies are two contracts, and they can still be one path: they become one wherever they converge, on the same long-haul route, in the same conduit, through the same building entrance or onto the same rooftop. Providers call all four kinds of separation below “diverse.” Only a diverse route keeps you online when the fiber is cut, and it is the one that takes work to confirm.
- Diverse provider
- Different companies bill the two circuits. It is the weakest of the four, and the one most often sold as the strongest.
- Diverse equipment
- The circuits terminate on different hardware in your comms room, which protects against a failed router and nothing outside the building.
- Diverse entrance
- The circuits come into the building through different rooms, so damage in one room cannot cut both.
- Diverse route
- The circuits share no physical infrastructure between your building and the point where traffic reaches the wider internet.
What to ask each provider
Put the questions in an email, so the answers exist in writing.
- Which entrance does the circuit use, and which riser does it climb?
- Which street route does it take to the first aggregation point?
- Which long-haul route does it ride after that?
- Where do our two circuits converge upstream?
Some providers answer all of it. Others answer part and decline the rest on security grounds, and a flat refusal tells you something as well. What you want to learn is whether anyone at the provider has traced the route. A “yes, they are diverse” with no route behind it tells you no one has.
Why the second path we build is a radio link
When a provider cannot say where its circuit runs, you can still choose a second path that cannot run there. A radio link cannot share a conduit with a wireline primary, because it uses none, so a radio link is what we build.
No conduit to share
The link travels through the air between our network and your building, so it shares no conduit, street trench or building entrance with a wireline circuit.
Upload that fails over too
The link is symmetrical, so when traffic moves onto it, the backups, calls and remote sessions leaving your building move with it.
Capacity for your site alone
If the backup has to keep its capacity for your site through a regional outage, ask for it as a dedicated circuit.
What automatic failover needs from the equipment you own
We deliver the second circuit. Whether your traffic moves onto it by itself depends on four things inside your building.
Something that decides
Failover is a decision, and a device has to make it without a person. That is either a router running BGP with address space you control, or a policy on the edge device that tests the primary continuously and moves traffic when the test fails. A link that shows as up while it drops traffic is the case that catches most designs out.
Addressing and DNS
If your public services live on the primary circuit’s addresses, failing over changes the address, and cached DNS records keep pointing at the old one until their time to live runs out. Either carry portable address space across both paths, or shorten the records that matter and find out which resolvers ignore that.
Sessions in progress
What survives the cutover depends on where each session terminates. Traffic translated onto the failed circuit’s address drops and has to reconnect. Calls anchored at a hosted platform usually re-register within seconds, though a call already in progress drops. Decide in advance which of those matters enough to engineer around.
A testing cadence
A backup path that has never carried production traffic is a hypothesis. Fail over to it on purpose, in a window you choose, often enough that the first outage is not the first time anyone watches it work. Write down what broke and fix it, because something usually does on the first attempt.
Backup internet and failover questions
The edge device sees the primary fail a check it was already running, withdraws that path or lowers its priority, and traffic follows the remaining route. How long that takes depends on your own configuration: detection intervals, routing timers and how quickly an upstream neighbor accepts the change. A cutover time quoted without knowing your edge setup comes from a datasheet. What tells you more is which of your applications noticed the switch.
For a small office whose backup carries email, card payments and a few calls for an afternoon, cellular is a reasonable and inexpensive answer. For a site running hosted applications, replication or a contact center it usually falls short, for three reasons: the capacity is shared with everyone else on the tower, the addresses are usually not yours, so public services do not follow you, and the regional event that damages fiber can load or knock out the cellular network at the same time. A site that cannot go quiet needs a second path with capacity of its own.
No. You can keep both and add our link as a third path, or replace whichever provider shares the other one’s route once the path review shows which one that is.
Tell us how your edge routes today, whether you hold your own address space and how many addresses the site needs, and we will answer with the quote.
We quote each link for its own building once the survey is done. Contract and no-contract terms are both available, so ask for the price under each.
Have us trace the path you already pay for
Send us the two providers you have and the building address. We will tell you where we think the paths converge and what an independent second path would look like at that site, including when you do not need one.