Design and deploy a link-bonding algorithm for resource-constrained edge networking devices that evaluates multiple WAN paths in real time and steers traffic based on latency, loss, jitter, cost, and availability — without exceeding the compute and memory limits of ARM-class CPE hardware.
Link-Bonding IP in Production Networking Devices
OBJECTIVES/GOAL
CHALLENGES
Running a per-flow scoring and steering engine on devices with 256 MB to 1 GB RAM while maintaining low-latency forwarding across Ethernet WAN, Wi-Fi, LTE modem, and VPN tunnel paths. Adapting the algorithm to handle simultaneous traffic classes with different failover requirements from a single policy engine.
ACCOMPLISHMENTS
Benison has deployed the link-bonding algorithm across thousands of field devices at remote sites with unreliable access links. The algorithm handles per-flow link decisions, failover, and duplication entirely in software on embedded Linux / OpenWrt-class runtimes. The robotics adaptation maps the same policy engine to robot traffic classes: safety/control, teleop, telemetry, OTA, logs, and video snapshots.
TECHNOLOGIES USED
ARM edge gateway / CPE class hardware
Embedded Linux / OpenWrt-style runtime
Multi-link bonding: Ethernet WAN, Wi-Fi, LTE, VPN tunnel
