Bonding lab

One connection. Many paths.

Throughput nowLarge file raceSessions droppedCall silenceStatus

Conventional 0Mbps

0 0.0s All links up

Bonding 0Mbps

0 0.0s Tunnel up

How to read the diagram

Same links, same people, two ways to use them

Conventional failover

A multi-WAN router gives each session one link, weighted by link speed. Each link reaches the internet with its own public address. When a link fails, the router keeps sending into it until three health checks fail, then the sessions on it reconnect over another link with a different address. Calls rejoin, uploads restart, and the large file starts again from zero.

Connection bonding

The bonding router spreads every session's packets across all available links inside one encrypted tunnel to the bonding hub. The Hub puts the traffic back together and presents one address to the services. When a link fails, traffic moves to the remaining links in a fraction of a second and the sessions carry on with less capacity.

Click a link's label in the diagram, or untick it in the settings, to cut it. Cutting the busiest link shows the biggest difference. Restoring a link shows another one: conventional sessions stay where they were moved, while bonding uses the restored link straight away.

What is modelled

Illustrative model, not a product benchmark

Router capacity: bonded throughput is capped at 2,000 Mbps, the vendor's published bonding capacity for all three models. The Hub is assumed not to be the bottleneck. None of the scenarios reaches this cap.
Hardware: the desktop model is a router with one built-in 2.5 Gbit/s Ethernet WAN port and three hot-plug module slots (4G, 5G dual SIM, VDSL or Ethernet), so up to four links. The mobile model has three 4G LTE modems with two SIM slots each (one active, one standby) and Ethernet WAN ports for other links such as Starlink. The rack model has seven Ethernet WAN ports and no modems.
Link speeds in the scenarios are typical examples, not measurements. Set them to match your own lines.
Demand per person: video call 2.5 Mbps each way; live video 8 Mbps up or 5 Mbps down; web and email 0.3 Mbps up or 2 Mbps down; cloud sync takes what is free, up to 5 Mbps up or 20 Mbps down each. The large file takes whatever is left.
Sharing: each link is shared fairly between the sessions on it: nobody gets more than they need, and nobody is starved while the link has capacity. If a link is overloaded, calls and video slow down along with everything else. Bonding loses the tunnel overhead set in the assumptions (5% by default); conventional links lose nothing.
Failure: conventional detection takes 3 s, then reconnection takes a further 6 s for calls, 5 s for live video, 4 s for uploads and 1.5 s for web. The large file restarts from zero. Bonding stops using a failed link after 0.3 s.
Race speed: the large-file bars run faster than real time so a race finishes in about half a minute. The times shown are real transfer times.
Latency, jitter, packet loss, congestion control and the vendor's own scheduling are not simulated. Real results depend on the configuration and the lines.

Addresses shown are from the documentation ranges (203.0.113.0/24 and 198.51.100.0/24) and do not belong to anyone.

References

About this piece

Same links, same people, two ways to use them

Conventional failover

Connection bonding

Illustrative model, not a product benchmark