Budgeting Bytes: Acing Cost-Efficient Video Storage
summary
Vimeo discloses a concrete system: a policy calculator that models storage and retrieval costs across GCS tiers and an ML classifier that labels each video hot or cold using CDN and origin logs, autoML and custom features and a light GBM model. The system cuts storage bills ~60% with the calculator and yields an additional ~30% reduction when applied per‑video, trading occasional retrieval penalties against massive sustained savings and iterating via error analysis and custom feature engineering.
AI-written summary · about
chapters (10)
- 0:00Welcome
- 1:06Video storage costs
- 3:14Storage tiers primer
- 5:22Policy calculator
- 8:53ML for storage classification
- 12:57Feature engineering
- 15:04Ground truth labeling
- 16:49Model performance
- 18:37Model improvements
- 20:00Summary & Q&A
related talks
18mOptimizing Video Storage Via Semantic Replication
24mOptimizing Cloud Costs with Machine Learning - A CloudVane Story
30mMLDS 2019: Anubhav Shrivastava of Voot at Machine Learning Developers Summit 2019
01h 04mBuilding Efficient and Frugal Image Data Processing Pipelines
12mRevolutionizing Data Management
24mAI-Assisted Media Planning
17mContent Aware Encoding for low latency live streaming encoders using deep learning
33mArchitecting Scalable Multi-Modal Video Search