Keeping a site online when a link goes down — and what failover cannot do
Failover is the cheapest way to stop a site going dark when one provider fails. It is also frequently the wrong purchase, because most Indian last-mile links degrade rather than die, and failover barely reacts to that. Here is how it works, where it falls short, and how to tell which mode you actually need.
Internet failover means a site keeps a second connection in reserve. A router watches the primary link, and when it judges that link to have failed, it moves traffic across to the backup. When the primary recovers, traffic moves back. For most of the day the second connection sits unused, which is the cost of the arrangement and also its simplicity.
It is the cheapest form of redundancy and it solves one specific problem well: a site that would otherwise go completely offline when a single provider has an outage. If your concern is "the office lost internet for three hours last Tuesday", failover addresses it.
What it does not do is make a connection faster, and it does not keep sessions alive across the switch. Those are different problems needing different mechanisms, and conflating them is the most common reason people are disappointed by a failover purchase.
Failover assumes a link either works or does not. On Indian last-mile connections that assumption is often wrong. A cellular link at the edge of coverage, or a broadband line during evening congestion, does not stop answering — it just becomes unusable. Loss rises, latency becomes erratic, throughput drops to a fraction of what it was.
To a router checking whether the interface is up, that link is healthy. It keeps sending traffic down it. The site is effectively offline while every indicator reads normal, and nobody gets an alert because nothing failed.
This is why the relevant question when buying is not how fast a device switches, but how it decides something is wrong. A device that probes a remote target and watches loss and latency will catch a degrading link. A device watching only whether the modem is connected will not.
These terms get used interchangeably in product listings, which makes comparison difficult. They describe genuinely different behaviour, and the difference decides whether a device solves your problem.
The practical test is what happens to a single connection — one video call, one file upload, one VPN tunnel. Under failover it survives only if no switch happens. Under load balancing it is pinned to one link and limited to that link's capacity. Under bonding it uses every link at once and keeps running when one of them degrades.
Failover is the right choice when your traffic is ordinary browsing, email and cloud applications, when a brief interruption during a switch is acceptable, and when one of your links is genuinely good most of the time. It is the cheapest option and the simplest to run.
Bonding earns its extra cost when sessions must not break, when you are pushing video or large uploads outward, or when no single link at the site is adequate on its own. Live broadcast, telemedicine, remote surveillance and field operations tend to fall here — not because they are more important, but because their traffic is continuous and outbound.
A useful way to decide is to ask what breaks when a link goes bad for thirty seconds. If the answer is "someone reloads a page", buy failover. If the answer is "we lose the stream" or "the upload restarts", the extra cost of bonding is what you are actually buying.
Specification sheets in this category are written loosely, and several of the things that matter most are rarely stated. These are worth asking any vendor directly, including us.
What happens to one continuous session — a call, a stream, a large upload.
| Failover | Load balancing | Bonding | |
|---|---|---|---|
| Links in use at once | One | All | All |
| A single session can use | One link | One link | All links together |
| When a link fails completely | Switches; session usually drops | Session on that link drops | Session continues |
| When a link degrades but stays up | Often does not react | Keeps using the bad link | Traffic shifts away from it |
| Effect on upload capacity | None | More total, not per session | More per session |
| Relative cost | Lowest | Moderate | Highest |
Specifications below are drawn from the current product datasheets. Ask us for the datasheet of any device before you commit.
| Device | Links | Suited to | Notes |
|---|---|---|---|
| Clover 2X | Dual 4G modems | Fixed sites — offices, branches, remote installations | Load balancing and automatic failover. One gigabit WAN and four gigabit LAN ports, metal IP30 enclosure, −20°C to 60°C, rated for 24/7 operation. |
| Zifilink 5X | Up to 5 USB cellular dongles | Live video and high outbound bandwidth | Bonds up to five cellular networks. Wi-Fi hotspot plus one LAN output. Zero configuration — detects and connects on power-on. |
| Bloom Tulip | 4 × 4G modems | Field operations away from mains power | Up to 3 hours on internal battery with V-mount battery support for extended use. One RJ45 Ethernet, 802.11a/g/n/ac Wi-Fi, 684 g. |
| Clover M2 range | Varies by model | Vehicle-mounted and outdoor deployments | Several configurations for in-vehicle and outdoor use. Tell us the deployment and we will point you at the right one. |
Not sure which fits? Use the product finder or tell us the site conditions and we will recommend a configuration.
Internet failover is an arrangement where a site has more than one internet connection, and a router moves traffic to a backup link when the primary one stops working. The backup normally sits idle. Failover is about keeping a site reachable, not about making it faster.
Usually yes, briefly. Failover works by switching paths, and a switch changes the public IP address that traffic leaves from. Anything holding an open session — a video call, a VPN tunnel, a live stream, a large upload — generally has to re-establish. Short interruptions to browsing go unnoticed; real-time traffic does not.
Failover uses one link at a time and switches when it fails. Load balancing uses several links at once but sends each connection down one of them, so a single upload is still limited to one link. Bonding combines links into one logical path, so a single session can use all of them at once and survives a link degrading rather than only a link failing.
Because links here more often degrade than die. Packet loss climbs, latency spikes, throughput collapses — but the link still answers, so failover logic sees it as up and never switches. The connection is unusable while the router reports everything as normal. Bonding handles this case because it spreads traffic continuously rather than waiting for a failure event.
Yes, and it is usually a better arrangement than two connections from one operator. Distinct operators rarely fail at the same moment, and in most Indian locations coverage varies enough between operators that one will be usable when another is not. A dual-modem router such as the Clover 2X is built for exactly this.
Not for general internet access. You do need it, or an equivalent arrangement, if remote systems have to reach the site — for inbound VPN, remote monitoring, or site-to-site links. Benlycos devices support static IP where the deployment calls for it.
That depends on how the router detects a failure, and it is worth asking any vendor for their method rather than only a number. Detection based on probing a remote target is slower but more accurate than watching whether the physical interface is up, which misses the common case of a link that is connected but not passing traffic.
Our connectivity experts are ready to help you find the perfect solution for your business needs.
Talk to an Expert