Reducing Instagram’s basic video compute time by 94 percent

This post details a specific optimization within Instagram's video processing pipeline. By repurposing progressive encoding frames to create basic ABR encodings, Instagram reduced the compute cost of basic video encodings by 94%. This freed up resources, allowing for increased production of advanced encodings, which improved overall video quality and playback smoothness for users. A testing framework was developed to validate the net positive impact of this change by comparing traffic across control and test pools.