RenderStream Networking
Disguise RenderStream is a protocol and technology developed to enable real-time streaming of rendering data between systems, such as media servers and rendering engines.
It is designed for synchronised, high-quality, and low-latency workflows, commonly used in applications like:
- Virtual production
- Live events
- Immersive experiences
Bandwidth Considerations
Section titled “Bandwidth Considerations”- RenderStream handles high-resolution, high-frame-rate content, including textures, video, and rendering data.
- Uncompressed data significantly increases bandwidth demands.
Transport types
Section titled “Transport types”Streams in RenderStream can either be compressed or uncompressed.
Uncompressed streams
Section titled “Uncompressed streams”RenderStream Uncompressed Streams (sometimes shortened to RSUC) are low latency, high-quality, and require much higher bandwidth.
Uncompressed streams require specific hardware:
- Network Interface Cards (NICs): 25Gb or 100Gb Mellanox network interface.
- Multiple Machines: dedicated render nodes render and send content via uncompressed streams to actor and director machines.
- Genlock: required when rendering multiple fragments that are output by multiple actor machines.
Supported uncompressed formats:
- RGB 4:4:4 8bit
- RGB 4:4:4 10bit
- RGB 4:4:4 12bit
- YUV 4:2:2 8bit
- YUV 4:2:2 10bit
- YUV 4:2:2 12bit
The transport format can be configured in the cluster assigner settings of the workload: RenderStream Layer
Compressed streams
Section titled “Compressed streams”Compressed streams use H.265 compression to reduce bandwidth needs while providing consistent content quality. Compressed streams use an order of magnitude less bandwidth than uncompressed and can be used on the 1G, 10G, or 25G+ networks. A limited number of streams can be encoded and decoded by the hardware on the cards.
Supported compressed formats:
- YUV 4:2:0 8bit
The transport format can be configured in the cluster assigner settings of the workload: RenderStream Layer
Hardware Considerations
Section titled “Hardware Considerations”-
Practical Throughput:
- Even with 100Gb Ports, actual throughput is typically around 80Gb due to protocol overheads, error checking, and operational processes. Using the Assigner and predictions on workload traffic you can aim to calculate the expected throughput needed per workload.
-
Clustered Rendering:
-
Allows for the distribution of rendering tasks across multiple RX servers, improving performance and reducing latency.
-
Provides redundancy and fault tolerance, ensuring that if one machine fails, others can take over the workload.
-
Enables horizontal system scaling, allowing for the addition of more machines to handle increased workloads.
-
However, increases cluster management network traffic, which at very large scales can become a bottleneck.
By using a structured and distributed system like Fabric, high-bandwidth, low-latency workflows become more manageable as this is included in our system designs, especially in complex rendering scenarios.