Concurrency, Functional Paradigm and Dynamic Flow Control in Multimedia Processing
summary
Under-the-hood look at Membrane (Elixir) core shows a working RTMP sender/receiver pipeline demo in Livebook, with real measurements: a chain of 11 filters processing 200 ms per buffer yields 5 buffers/sec until one filter slows to 500 ms and throughput falls to about 2 buffers/sec then latency and memory climb. Backpressure fixes unbounded queues but can drop throughput to ~0.4 buffers/sec unless each element has an input queue of configurable size, and live sources are handled by turning backpressure off so streams keep flowing.
AI-written summary · about
chapters (9)
- 0:00Introduction to talk
- 0:43What are pipelines
- 3:32Pipelines in multimedia
- 4:39What is Membrane
- 6:51Actor model in Membrane
- 10:05Live demo of pipelines
- 12:21Solving pipeline problems
- 21:52Membrane core new version
- 22:57Q&A session
related talks
33mMembrane Framework - the true story of (...)
38mMembrane Framework: Bringing BEAM into multimedia
27mMembrane SFU: Bringing WebRTC to Elixir
31mManaging Latency in Continuous GNU Radio Flowgraphs
25mHow we built Algora.TV (livestreaming for developers) using Membrane and Elixir
28mPipelining WebRTC videoconferences in Elixir with Membrane Framework
27mBuilding a Low-Latency Voice Assistant leveraging Elixir and Membrane
49mThinking in a Highly Concurrent, Mostly-functional Language