Intelligent streaming, primed for live

Rilla allows your live streaming viewers to safely relay data using their upstream bandwidth, improving video stability and reducing load on your CDN. A lightweight SDK integrates into your streaming stack, while an AI orchestrator intelligently decides when and how viewers share data in real-time - always falling back to the CDN if needed.

This means Rilla customers can safely augment their existing CDN capabilities, saving saving 60-90% on CDN transport costs without compromising the viewer experience.

We differentiate from historical P2P solutions in several ways:

Traditional P2P models were focused on static file transfer; whereas Rilla is engineered for the dynamic and unpredictable demands of real-time video streaming 

Key differences are: 

• Real-time orchestration: Rather than using basic trackers/swarming, we use real-time orchestration based on dynamic network conditions. This allows us to handle significant network volatility without disruption to the viewer experience
• Stable high-thoughput quality: All peers are not treated equally, If a peer loses the ability to contribute, that node will immediately and safely fall back to CDN
• Integrated for streaming: We act as a fully integrated layer that handles key concerns including adaptive bitrates and content protection technologies that previous P2P solutions largely ignored

Rilla reduces content distribution costs by safely deflecting traffic away from your CDN using the users available upstream bandwidth.

Typically, in a network where 30% of the audience can consistently contribute bandwidth during the live event, we can achieve 80-90% CDN deflection.

This results in significant cost savings on the traffic that would otherwise be served by the CDN.

Rilla uses an AI-orchestrator service that observes network conditions to optimize video delivery between peers.

We use deep reinforcement learning (DRL) to train the system on billions of complex and dynamic network conditions.

During training, DRL agent learns an optimal, real-time policy for complex scenarios - such as peer selection, load distribution, and CDN offload ratios.

During live streaming, those learned models run at a fraction of the cost, enabling Rilla to orchestrate the network with predictive accuracy and high reliability.

Yes, It improves, especially at scale. 

Our AI-orchestrator is trained to maintain increased integrity, stability and reduce rebuffering at scale..

The larger the concurrent audience of the P2P network, the more inherently stable it becomes.

This has a network effect meaning the more viewers that are connected to a reliable source of traffic, the better experience.

With continuous monitoring and real-time orchestration, every viewer’s upstream capacity and network conditions are managed in real-time. P2P networks can be unstable as we rely on the upstream bandwidth of each contributing node. If local network conditions or contention become an issue then we need to accommodate this. Similarly if a local power outage occurs or some fans elect to quit en masse then the Rilla orchestration technology instantly adjusts the topology of the network within “buffer time” to ensure that Quality of Experience is maintained.

Rilla's architecture is engineered for massive, dynamic audiences through horizontally scalable orchestration.

How Rilla Scales:

• Massive Horizontal Scaling: Rilla's backend is designed to scale horizontally up to multiple millions of viewers within regional boundaries.The AI orchestrator is implemented on a proprietary,designed for coordination-free operation.
• Network Effect: Every Rilla-enabled viewer strengthens the network, automatically joining a regional relay layer. This expands delivery capacity as the audience grows, allowing streams to scale beyond the limitations of traditional CDNs.
• Low-Latency AI: Rilla's AI model is optimized for low-latency inference, making peer topology decisions in mere tens of microseconds. It supports over 10,000 matchmaking decisions per second per shard.
• Dynamic Audience Scaling: The lightweight signaling service accommodates tens of thousands of new viewer connections per second, handling the most abrupt surges in audience size.