Latency you can defend
Low-latency HLS on a dedicated edge, with the ladder tuned to the material rather than to a default profile. Sport and studio content do not want the same ladder.
Linear, live and a back catalogue on one pipeline
A broadcaster's traffic is not one shape. A cricket match and a fifteen-year drama archive have opposite cache behaviour, opposite failure tolerance and opposite cost profiles — and most platforms are built for exactly one of them.
Who this is for
Organisations whose traffic has a scheduled cliff in it, and whose worst night of the year is known months in advance.
- Broadcasters
- Terrestrial, satellite and cable operators taking linear channels and a back catalogue to streaming on one pipeline.
- OTT platforms
- Services carrying live rights alongside on-demand, where a match and an archive compete for the same delivery capacity.
- Sports rights holders
- Organisations streaming events directly, whose concurrency is decided by a fixture list rather than by a growth curve.
Peak night is the architecture
An average-day design fails on the one night that matters. Broadcast concurrency is not a smooth curve — it is a flat line with a cliff in it, and the cliff arrives on a schedule everybody already knows about. That is a gift, because it can be planned for, and a trap, because a platform sized for the average will fall over exactly when the audience is largest and the coverage is widest.
The structural answer is separation. Live gets its own edge, its own cache policy and its own failure domain; the on-demand catalogue gets another. A surge on the live path then cannot evict the archive's working set, and an archive re-encode cannot compete with a match for delivery capacity.
What a broadcast deployment has to get right
Channels that behave like channels
Scheduled playout, ad markers, blackout and regional variation — the operational vocabulary a broadcast team already works in, rather than a VOD system pretending.
A back catalogue that is cheap to hold
Most of an archive is watched rarely. Storage class and cache policy follow that reality instead of treating a 2009 episode like tonight's match.
Window enforcement
Rights are the constraint that most often turns into an outage. Licensing windows are applied per title and re-checked at the edge on every segment rather than trusted to the client. Country-level blocking is a separate control: the edge resolves and logs the viewer's country today, and enforcing a deny list on it is build work, not configuration.
Live and VOD, deliberately apart
The separation is the point. Two edges, two failure domains, one control plane.
Untrusted — the public internet
- Web player
- Mobile apps
- TV apps
- Contribution
TLS terminates here and nowhere else
- traefik-edge
- CrowdSec
- Live path
- VOD path
Reachable only through the edge, or through the bastion for operators
- app-core
- content-api
- search-api
- session-authz
- billing
- key-delivery
- recommendation
- bastion
A match-night surge cannot evict the back catalogue
- live-transcode
- live-origin
- vod-transcode + GPU
- vod-origin-cache
- upload-api
- job-scheduler
- RabbitMQ
- MinIO
State lives inside the perimeter
- MariaDB
- Redis
- Elasticsearch
- ClickHouse
- Redpanda
- OpenBao
Declared entirely in code — the estate rebuilds from the repository
- Proxmox VE
- Terraform
- Ansible
- Pipeline → registry
- Prometheus
- SigNoz
- Grafana
- Loki
The things broadcasters ask first
Can this take our existing playout feed?
Yes. The ingest path accepts standard contribution feeds and the encoding ladder is built around your material. The platform sits downstream of playout, not in place of it.
How do you handle a night that is ten times normal traffic?
By sizing the live edge against your real peak and keeping it separate from the catalogue, so the surge has somewhere to go that does not damage everything else. Capacity for a known event is a planning problem, and broadcast peaks are known.
What about advertising?
Ad markers are carried through the pipeline and the insertion integration depends on which ad system you already run. That integration is scoped in discovery rather than assumed.
Tell us about your peak, not your average
The number that decides a broadcast architecture is the concurrency on your biggest night of the year. Send that, the channel count and the archive size.