Asynchronous Sharing of Media Essence Data in Software Defined Workflows 

As media workflows migrate to open-source software-defined frameworks consisting of a collection of containerized applications running as processing nodes on a cluster of commercial-off-the-shelf (COTS) servers, there is a need for these distributed application processes to asynchronously share media essence data in the most efficient and secure way possible without going back to the baseband transport. This sharing of media essence data is complicated by the fact that some applications perform graphics processing unit (GPU) processing requiring the essence data to be in GPU memory while others perform media processing on the central processing unit (CPU) requiring the essence data to be in system memory. Several technologies exist for such sharing of media essence data between processes on different compute nodes of a cluster so the ultimate solution needs to support multiple transports with an application layer API that support initiating and terminating communications. We compare the various techniques and demonstrate how they can be used within a platform available today for the easy development and deployment of distributed media process pipelines for transcoding and AI applications.

Gareth Sylvester-Bradley | NVIDIA Development UK Ltd | Reading, United Kingdom
Pravin Sethia | NVIDIA Graphics Pvt Ltd | Pune, Maharashtra, India
Thomas True | NVIDIA Corporation | Santa Clara, Calif., United States