Network-Scale Orchestration

Abstract

We describe the orchestration layer that balances intake, line assignment, and throughput across the Jones Foods autonomous network, executing network-wide rebalancing decisions in under one second.

The problem

Intake volume is weather-dependent, transport-dependent, and seasonal; line capacity shifts with maintenance windows and equipment generations; and the cost of a wrong assignment compounds down the line. The scheduler solves a bin-packing problem whose shape changes every 400 milliseconds, at network scale.

Architecture

We run a three-layer hierarchical scheduler — network, facility, line — with each layer planning on a different horizon (hours, minutes, seconds). The layers are permitted to disagree: the network planner is allowed to be wrong about the next ten minutes as long as it is right about the next ten hours, and lower layers repair local error without global replanning. Rebalancing decisions propagate network-wide in under one second over a dedicated telemetry mesh.

Results

Lines run in bursts as intake arrives; with no shifts to plan around, idle line-hours cost nothing but capital, and the scheduler exploits that freedom aggressively.