Encoding, Fast and Slow - Low-Latency Video Processing Using Thousands of Tiny Threads
Explore a groundbreaking video processing system called ExCamera in this 23-minute conference talk from USENIX NSDI '17. Discover how researchers from Stanford University, UC San Diego, and MIT developed a framework for low-latency video editing, transformation, and encoding of 4K and VR content using cloud function services. Learn about the system's two major contributions: a parallel computation framework utilizing thousands of threads, and a video encoder designed for fine-grained parallelism. Gain insights into the key principle of dividing video encoding work into fast and slow components, enabling efficient parallel processing. Understand how this innovative approach can revolutionize video processing techniques and improve performance in cloud-based environments.