The Reality of Industrial Networks
Industrial networks are not datacenter networks. Connectivity drops during shift changes when equipment powers down. VPN tunnels between plant and cloud fail during ISP maintenance windows. Firewall rules change without notice. WiFi coverage in a metal-heavy manufacturing environment is inconsistent.
Any architecture that assumes reliable connectivity will lose data. This is not a theoretical concern - it is a weekly occurrence in most facilities.
Store-and-Forward Architecture
Durable buffering writes every data point to local persistent storage before attempting delivery. If the destination is unreachable, data accumulates locally with configurable retention policies. When connectivity is restored, buffered data replays automatically in the correct order.
This is not an optional feature. It is the minimum requirement for production industrial data systems. Without it, every network event creates a gap in historical data that cannot be recovered.
Ordered Replay
Replay must preserve the original ordering and timestamps. Destinations receive a complete, ordered history - not just current state. This matters for:
- Historical analysis that depends on temporal sequences
- ML training pipelines that require complete datasets
- Compliance requirements that mandate data completeness
- Process correlation that spans network outage windows
The data plane must guarantee: no data loss, ordered delivery, and complete replay. Anything less is not production-ready.