A second Internet circuit does not automatically create redundancy. If both links share a carrier, physical route, device, or a router that checks only port status, they can still fail together. A dependable solution removes shared failure points, detects “link up but Internet down,” switches automatically, and is tested regularly.
When does a business need backup Internet?
Measure impact rather than employee count. Redundancy is worthwhile when an outage stops POS, electronic invoicing, IP telephony, video meetings, VPN, cloud cameras, SaaS applications, or customer support. Estimate one hour of downtime through lost sales, idle labor, repeated transactions, and reputational damage.
An office using only non-urgent email may tolerate a manual hotspot. A retail site or cloud-dependent operations team needs automatic failover. Higher availability also requires backup power, monitoring, and an incident procedure.
Three practical deployment models
| Model | Good fit | Trade-off |
|---|---|---|
| Primary fiber plus 4G/5G backup | Small office or shop | Simple, but bandwidth, CGNAT, and data allowance may be limited |
| Two fixed lines from separate ISPs | VoIP, many users, heavy traffic | More stable, but physical path diversity must be verified |
| Two WANs plus multi-path SD-WAN/VPN | Multiple branches or critical applications | Greater control with higher cost and operational complexity |
For many small businesses, fiber from ISP A and 4G/5G from carrier B is a sensible start. Locations with weak mobile coverage should consider another fixed circuit or a properly installed external antenna.
1. Remove common failure points
Two contracts may still use the same infrastructure. Ask about the last mile, aggregation point, and building entry. Check whether cables share a conduit or communications cabinet. For cellular backup, choose a carrier different from the primary ISP and survey signal during busy hours.
- Use a separate modem or ONT for each WAN.
- Connect them to independent WAN ports on the gateway.
- Put the ONTs, cellular modem, router, switch, and access points on a UPS.
- Place cellular equipment where reception is strong and away from metal enclosures.
- Do not let one outlet, adapter, or intermediate switch become the single point of failure.
Internet redundancy without backup power for the ONT and router solves only half the problem.
2. Select a router by capability, not port count
The gateway should provide at least two WAN interfaces, active health checks, failover and failback, NAT, and firewall policies on both links. Evaluate throughput with IPS, VPN, QoS, and filtering enabled; raw routing figures can be much higher than real-world secured performance.
Useful features include event logs, alerts, policy-based routing, backup-link bandwidth controls, DNS continuity, and secure remote administration. For inbound access, examine public addressing, CGNAT, port forwarding, VPN, and DDNS behavior on each circuit.
3. Failover or load balancing?
Active/standby sends all traffic through the primary WAN and activates the backup only on failure. It is predictable, suits metered cellular service, and reduces problems for applications sensitive to source-IP changes.
Load balancing distributes new sessions between links according to weights. It increases aggregate capacity for many users but does not combine both links into a faster single download. Session stickiness normally keeps each flow on one WAN.
VoIP, payments, VPNs, and IP-allowlisted services are often more stable with failover or a fixed routing policy. Guest Wi-Fi, updates, and less-sensitive traffic can use the second link to gain capacity.
4. Health checks must look beyond the gateway
A connected cable or successful ping to the default gateway proves only the path to the ISP device. The provider may still have a routing, DNS, or upstream failure. Probe multiple stable Internet destinations through each WAN using ICMP, DNS, or HTTPS as supported.
- Use at least two or three targets on separate networks.
- Declare failure only after several consecutive failed probes.
- Require a recovery threshold and stable period before failback.
- Track latency, packet loss, and jitter as well as up/down state.
- Avoid a target that blocks probes or depends on your own infrastructure.
Failing over too quickly causes flapping; waiting too long extends downtime. Begin conservatively, measure, and tune. MikroTik's documentation also demonstrates multiple probe targets to avoid letting one host trigger a false decision.
5. Define backup-link policy
4G/5G often has lower or less predictable capacity and may have a data cap. During failover, prioritize POS, DNS, identity, email, VoIP, and business applications. Temporarily restrict cloud backup, operating-system updates, high-resolution video, and guest Wi-Fi.
Use QoS for measured needs, but do not expect it to repair a saturated circuit. Shaping slightly below measured uplink speed lets the gateway manage queues. Choose a conservative value for variable cellular performance.
6. Plan for public IP changes
Failover usually changes the public source IP, so active TCP sessions, calls, VPNs, or logins may reconnect. That is normal for basic dual-WAN and is not truly seamless session continuity.
- Use rapidly reconnecting VPNs or outbound-initiated tunnels.
- Allow both addresses when a SaaS platform uses an IP allowlist.
- Update DNS or DDNS for inbound services while respecting cache time.
- With cellular CGNAT, prefer outbound tunnels instead of port forwarding.
- SIP and VoIP may favor failover over load balancing for consistent paths.
If an application must retain the same IP and live session during a circuit change, investigate an SD-WAN bonding, overlay, or managed-service design. Ordinary dual-WAN NAT does not provide that guarantee.
7. DNS, VPN, and internal services
Clients and the gateway need DNS that works through either WAN. Do not rely exclusively on a resolver reachable only through the primary circuit. For site-to-site VPN, build tunnels over both uplinks or document exactly how they re-establish after failover.
Office-hosted services need separate analysis. A CGNAT backup can provide outbound Internet while preventing inbound customer access. Public websites, email, and customer applications are generally better hosted in a cloud or data center than behind an office circuit.
8. Roll out in stages
- Measure current WAN bandwidth, latency, and peak utilization.
- Inventory critical applications, inbound requirements, and IP allowlists.
- Select a backup from a different carrier and, where possible, physical path.
- Install the dual-WAN gateway and UPS without enabling automatic failover yet.
- Test each WAN independently, including DNS, NAT, VPN, and firewall behavior.
- Configure health checks, active/standby mode, and failback delay.
- Apply QoS or backup-link traffic restrictions.
- Enable alerts and perform a controlled failover exercise.
- Document configuration, ISP contacts, and manual switching steps.
9. Test instead of assuming
Use a communicated maintenance window. Unplug the primary WAN to test physical failure, then leave the link connected while blocking upstream reachability to prove the health check detects an Internet failure. Measure the time from failure to restored user access.
While using the backup, test POS, calls, VPN, video meetings, DNS, cloud applications, and alerts. Restore the primary and observe whether failback causes another interruption or flapping. Safely simulate a power loss and confirm UPS runtime protects the whole network chain.
A useful exercise produces evidence: detection time, recovery time, interrupted applications, and corrective actions.
10. Monitor and operate
Alert when a WAN changes state, packet loss rises, cellular signal degrades, or the data allowance approaches its limit. Even when failover is invisible to users, operations must open an ISP ticket and avoid running unknowingly on backup for days.
- Exercise failover quarterly and after firmware upgrades.
- Review the SIM, data plan, renewal, and UPS condition monthly.
- Back up the gateway configuration after every change.
- Run supported firmware and limit administrative access.
- Review logs for recurring transitions or gradual degradation.
Estimate the full cost
Total cost includes the gateway, modem, antenna, UPS, installation, monthly circuit, SIM or data, monitoring, and administration. Compare that amount with expected outage loss. The most expensive appliance is not always required; a simple, tested architecture is generally more dependable than an unmonitored complex one.
Acceptance checklist
- The WANs avoid the same carrier or physical route where possible.
- ONTs, modems, gateway, switch, and access points have UPS protection.
- Health checks use several Internet targets, not only link or gateway state.
- Failover and failback thresholds prevent flapping.
- Critical applications receive priority on the backup circuit.
- CGNAT, public IP, VPN, and allowlist requirements are handled.
- Alerts work for WAN state changes and low data allowance.
- Recovery time was measured for both cable and upstream failures.
- Configuration, ISP contacts, and procedures are stored off the gateway.
Conclusion
Reliable backup Internet is a system of independent access links, a capable gateway, trustworthy health checks, backup power, traffic policy, and regular exercises. For a small business, primary fiber plus 4G/5G from another carrier often delivers excellent value. As requirements grow, the design can evolve to two fixed circuits, policy routing, or SD-WAN without changing these fundamentals.
No comments yet. Be the first to share your thoughts.