SpeedFusion 101 tells you what bonding does. This page is for the engineer who asks how — the packet flow, the ports, the overhead, and, just as important, the cases where bonding is the wrong tool.
The packet flow
SpeedFusion is a tunnel between two endpoints. Outbound traffic hits your Peplink, is sliced into tunnel packets, and each packet is scheduled onto one of the member WANs based on live per-path measurements (latency, loss, jitter, capacity). The far endpoint re-orders and reassembles, and hands your application one stable stream. The application never sees the member links — it sees the tunnel.
Endpoints come in three flavors: another Peplink device (site-to-site), a FusionHub virtual appliance you run in your own cloud or datacenter (a free Solo license covers one peer), or Peplink’s hosted service. Terminology matters: SpeedFusion is the engine built into the device — perpetual, no subscription. SpeedFusion Connect (SFC) is Peplink’s hosted endpoint service — quotas, renewals, and data limits you see in marketing apply to SFC access, never to the SpeedFusion engine itself.
Ports, encryption, and MTU
- Handshake: TCP 32015 to the remote peer; data: UDP 4500 by default (configurable per profile — the “data port” setting). If a network blocks outbound UDP, the tunnel can’t perform; fix the firewall, don’t fight it.
- Encryption: AES-256 per profile, or off for maximum throughput across paths you already trust (e.g., inside a private APN). Devices publish separate clear vs. AES throughput specs — check the datasheet for both numbers.
- MTU: tunneling costs header room. If you run another VPN inside SpeedFusion, clamp MSS and expect double-encapsulation overhead. Usually the better design: terminate the old VPN and let SpeedFusion’s AES be the encryption, or run the IPsec concentrator behind the FusionHub.
The overhead, in real numbers
Bonding is not free. Budget for it honestly:
| Mode | Typical overhead | Use when |
|---|---|---|
| Baseline bonding | ~14–18% | General traffic — the default |
| FEC (low / high) | ~13% / ~27% | Lossy links; rebuild lost packets without resends |
| Adaptive FEC | ~6–20% (floats) | Loss varies through the day — pay only when needed |
| WAN Smoothing | +100% to +400% (duplication) | Live video / voice where a glitch is unacceptable — capacity traded for certainty |
| Boost | ~10–15% per link | Throughput optimizer for lossy, high-latency bonds (cellular + Starlink) |
Failure modes — when bonding hurts
- Wildly asymmetric paths: one 900 Mbps fiber + one 5 Mbps LTE won’t make 905. Dynamic Weighted Bonding keeps the slow path from dragging the fast one, but the honest gain is resilience, not speed. Consider hot-failover instead of full bonding there.
- Large latency spread: mixing a 20 ms path with a 600 ms path forces re-order buffering. Smoothing/FEC help; physics still applies.
- Small-packet voice at scale: per-packet tunnel overhead is proportionally largest on tiny packets — size trunks accordingly.
- Single TCP flow on clean links: a lone TCP stream on one excellent path can beat the bonded tunnel (ramp-up + overhead). Bonding shines on aggregate, mixed, or lossy traffic.
Steering matters as much as bonding: a tunnel does nothing until outbound policy sends traffic into it. Start with the SpeedFusion 101 concepts, then build it in the User Guide.