Marcin Kolny discusses how Amazon's Prime Video team revamped its video quality analysis (VQA) service to reduce costs and improve scalability. Initially, the VQA team used a distributed system consisting of several components, such as media conversion, defect detectors, and orchestration management.
- The team re-architected the infrastructure, packing all components into a single process to eliminate the need for intermediate storage for video frames.
- The new infrastructure allowed the team to rely on scalable Amazon EC2 and Amazon ECS instances for deployment, resulting in cost savings.
- VQA initially used serverless microservices components, which raised scaling bottlenecks of 5% and cost concerns.
- To address these, VQA moved to a monolithic system, lowering infrastructure costs by more than 90% and increasing its scaling capabilities.
- In the new system, video frames are transferred in-memory, eliminating the need for Amazon Simple Storage Service (S3) as intermediate storage for video frames.
- This model also allowed VQA to monitor all live streams in real-time, resulting in a better customer experience.
- The shift to Amazon Elastic Compute Cloud (EC2) and Amazon Elastic Container Service (ECS) also allowed Prime Video to use Amazon EC2 compute-saving plans that lowered costs even further.
