Whitepaper

Cloud Video Platform: Realizing broadcaster-grade Cloud TV

How CVP delivers four 9s availability in the cloud

August 5, 2024

Download the whitepaper

Most broadcasters target at least four 9s of availability: 99.99%, or no more than 4.32 minutes of downtime a month. That's about one ad break. Many cloud services only commit to three 9s, which allows 43.2 minutes.

Moving TV to the cloud doesn't close that gap on its own. This whitepaper explains how the Cloud Video Platform (CVP) from Comcast Technology Solutions (CTS) reaches four 9s through a combination of engineering and operating discipline.

By the numbers: 99.99% availability target · 4.32 minutes of downtime a month, at most · 4 headline innovations · 3 more in development

What's inside

  • Why a four 9s cloud provider doesn't guarantee four 9s at the service level, and why mitigation time is the metric to watch
  • Multi-application fairness: three defenses against sudden traffic spikes
  • Failover: a geo-balanced active/active architecture with automated failover
  • Precomputation: how cutting runtime dependencies keeps one fault from cascading, with the Grant/Deny Service as an example
  • Business processes: the risk registers and runbooks behind a culture of reliability
  • What's next: chaos engineering, automated tooling and AI/ML

Frequently asked questions

Four 9s means 99.99% availability, which allows at most 4.32 minutes of downtime a month, about the length of one ad break. Three 9s, which many cloud services offer, allows 43.2 minutes a month.

No. The availability of the underlying cloud doesn't automatically translate into four 9s at the service level. When an issue occurs, what matters is mitigation time: how quickly the team detects, triages and restores service.

CVP uses multi-application fairness, with three defenses: preventing any one group of callers from monopolizing resources, responding with low-latency successes while overloaded, and rejecting only the requests that need overloaded resources.

CVP runs a geo-balanced active/active architecture on top of its cloud provider, so one active path can absorb traffic from a failed one. Custom anomaly detection triggers failover automatically, since manual intervention is too slow for four 9s.

Precomputation reduces runtime dependencies between systems, so a problem in one service is less likely to cascade. The Grant/Deny Service is an example: it precomputes entitlements so playback checks don't need to call other systems.

Please note, this whitepaper was last updated on August 27, 2025.

Download the whitepaper