Architecture - Video CDN [EN]

Video CDN was designed with a two-tier architecture and cloud-native storage principles. This design maximizes efficiency, performance, and cost-effectiveness for large-scale media delivery.

At the first tier, redirectors handle incoming client requests. Instead of serving content directly, they analyze each request and determine the most optimal delivery node based on a combination of factors — including geographic proximity, current load on storage and network resources, and delivery availability.

Redirectors respond with a redirect (302 Found) or a manifest (if HLS or DASH streaming was used) pointing to the appropriate edge node. Thanks to the relatively light traffic volume at this layer, anycast routing is used to provide built-in fault tolerance and global load balancing.

The second tier consists of edge nodes responsible for streaming content to end users. These hosts operate using unicast, which offers more flexibility in peering and bandwidth management. Nodes are capable of saturating uplinks up to 85–90% of their capacity, improving network utilization and helping reduce delivery costs.

Additionally, Video CDN uses intelligent, cloud-like storage replication. Files are not copied blindly to all servers. Instead, replication is driven by real-world demand — popular content is proactively replicated closer to viewers, while rarely accessed files remain centralized. This approach provides the benefits of distributed caching without the inefficiencies of full mirroring.

It’s important to note that the Video CDN is optimized for pre-imported content. In most cases, files must be imported in advance to unlock advanced features, such as multi-bitrate manifests or file clipping.

At the same time, the system remains flexible — it can also process requests for files that haven’t been imported yet, striking a balance between functionality and cost-efficiency.