If cloud applications, IP telephony and remote access depend on a single internet line, work stops when that line goes down. A second line reduces this risk; but putting a second modem on the shelf isn’t redundancy. Which infrastructure the lines come from, how the firewall will tell that a line really isn’t working, how internet-facing services will move to the backup line and how the failover will be tested must each be planned. An unplanned backup line is a bill whose uselessness is discovered at the first outage.
1. Real redundancy: different infrastructure, different route
If both lines come from the same carrier, the same exchange and the same cable duct, a single dig or exchange fault can cut both. For redundancy to be meaningful, the lines should come as far as possible through different carriers, different technologies (fiber, metro Ethernet, radio link, 4.5G/5G) and different building entries.
A different carrier alone isn’t enough either; in some areas different brands use the same infrastructure provider’s fiber. Before signing, ask the carriers which exchange and which route the line will come from. The same rule applies inside the building: if two fibers enter through the same manhole and the same duct, a single dig in front of the building cuts both.
A point that can be missed is power. The firewall is connected to a UPS, but the ONT, modem or radio link device often stays plugged into the wall socket. When the power goes out, the firewall runs but the lines don’t. All line termination devices should be fed from the UPS.
2. Failover or load balancing?
- Active–passive (failover): Traffic goes over the main line; when the main line goes down, the backup takes over. It’s simple, and the backup line can be a cheap mobile line.
- Active–active (load balancing): The two lines are used together; sessions are distributed across the lines. Total capacity increases, but a single session still uses a single line.
- Application-based routing (SD-WAN): Latency-sensitive traffic such as voice and video is sent over the best line, and heavy but flexible traffic such as backups over the other.
In an active–active setup, pay attention to the distribution method. If sessions are sent over a different line for each connection, the same user comes to the same site from one IP in one request and from another IP in the next. Banking portals, e-government applications and some SaaS products terminate the session in this case. Pinning by source IP (sticky) or a rule that defines a single line for these destinations is the solution to the “the internet works but the bank site kicks me out” complaint.
3. Health checks must be done right
A line appearing “connected” doesn’t mean the internet works; there can be a problem on the carrier side while the modem is on. On the firewall, define regular checks (ping, DNS or HTTP) to more than one external target for each line; set latency and packet loss thresholds. A check tied to a single target causes unnecessary line failover when that target doesn’t respond.
Choose targets from different networks, and consider a line down only when most of the targets don’t respond. Also verify that the check traffic leaves through the line it measures; with wrong routing, the second line’s check packet goes out the first line and the second line looks as if it never goes down. Use separate thresholds for going down and coming back. If the main line is flapping, switching every few seconds is worse than not switching at all. For failback, the line should be required to stay healthy without interruption for a certain time.
4. Internet-facing services and VPN
The most overlooked topic with dual lines is inbound traffic. If externally published services (remote access VPN, email, web, camera recording access) are tied to the main line’s IP address, these services become unreachable when the main line goes down. Options:
- Preparing DNS records and firewall rules for both lines
- Building IPsec tunnels between branches over both lines
- Defining a second connection address for remote access clients
- At the right scale, building a multi-carrier setup with your own IP block and BGP
In DNS-based failover, the record’s TTL is decisive; a TTL of hours means failover takes hours too. For branch tunnels, keeping both tunnels up and routing by priority (metric or administrative distance) gives faster results than waiting for a tunnel to be built during an outage.
There is one more detail with mobile backup lines: carriers mostly offer mobile lines behind CGNAT, without a public IP address. Such a line works for outbound traffic but accepts no connections from outside in. If the remote access VPN also needs to work on the backup line, a mobile line with a static IP or a second wired line with a static IP is needed.
5. Balancing capacity and cost
The backup line doesn’t have to match the main line’s capacity; but it should be decided which work must keep running during an outage, and the backup line must be able to carry it. With mobile backup lines, quotas and speed caps must also be taken into account.
The practical method is to draw up a priority list for the moment of an outage: ERP and accounting access, IP telephony, email and remote access first; backups, updates and video streaming last. Stopping cloud backups and operating system updates with a firewall rule while the backup line is active keeps the low-capacity line from clogging within minutes. If the quota runs out, the backup line itself gets cut in the middle of the outage.
6. Test failover regularly
A failover that has never been tested can surprise you at the first real outage. In a planned maintenance window, take the main line out of service in a controlled way; check internet access, VPN tunnels, IP phones and internet-facing services. Record the test and its result, and repeat it at regular intervals.
Don’t test only by shutting down the interface on the firewall. That method skips the health check; in a real outage the line appears “connected” but traffic doesn’t pass. A more realistic test is unplugging the modem’s or ONT’s WAN connection, or cutting traffic on the carrier side. During the test, note how many seconds the failover takes, which services recover on their own and what happens on failback. Failback is part of the test too. If NAT mappings and access rules also change when switching to the backup line, test them within the scope of firewall installation and rule management.
Checklist
- The lines come from different carriers and/or different physical infrastructure
- A multi-target health check is defined for each line
- A failover, load balancing or application-based routing model has been chosen
- Inbound services and VPNs are prepared for the backup line
- A failover test has been done and documented
Frequently asked questions
Is it enough if both lines come from the same carrier?
Often it isn’t. The same carrier and the same physical route can cause both lines to go down together in a shared fault. Where possible, a different carrier and different infrastructure (fiber, radio link, mobile) should be preferred.
Does load balancing speed up a single download?
Usually not. With session-based load balancing, a single connection uses a single line; the gain shows up in total capacity when there are many users and sessions.
Are open sessions preserved during line failover?
For general internet traffic, usually not, because the outbound IP address changes and the session is re-established. This should be tested for critical applications, and if needed, reconnection support on the application side should be checked.