Federation is the architecture that lets many separate sites and servers be managed and viewed as one system, without funnelling all their video to a central point. Each site records its own cameras locally and keeps running on its own even if the wide-area network drops; an operator at headquarters sees a single, unified view across all sites and pulls video from any of them on demand. It is how a retail chain, a city, or a multi-campus enterprise runs hundreds of sites coherently.
The defining pattern is record-local, stream-on-demand. Rather than streaming every camera continuously to a central data centre — which would need enormous bandwidth and create a single point of failure — federated sites store video where it is captured and send it upstream only when someone actually looks. A central management layer handles users, search, and health across the federation, so the system feels unified while the heavy video stays distributed; centralising would need around 10 Gbps where federation needs a fraction of that.
The key boundary, and the pitfall, is that ONVIF standardises the camera-to-VMS link, not VMS-to-VMS. Federation between servers is therefore vendor-proprietary, and joining different brands into one federation generally needs a higher-level integration layer (a PSIM or SDK work), not the open standard. Plan federation within a vendor's ecosystem where possible, design each site to survive a network outage autonomously, and treat cross-brand unification as a custom integration, not a given.
