How We Increased Image Search Throughput
summary
Describes an image search system architecture and current optimization techniques, comparing their pros and cons. Explains a tiered (multi-tier) indexing scheme—including an element nickname referencing atomic number 78 (platinum)—and shows how combining sharding with machine learning reduces hardware costs and improves traffic balancing and capacity.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 2:34Base search load
- 5:03Sharding and replication
- 7:53Boosting throughput methods
- 9:17Reducing document tops
- 11:06Degradation scenarios
- 12:12Query pre-classifier
- 13:58Platinum tier benefits
- 16:50Offline and online metrics
- 22:19Document classification
- 24:07Smart query routing
- 28:52Comparison with other methods
- 31:41Summary
- 32:44Q&A
related talks
53mHow Search Works
32mRetrieval Augmentation and Semantic Search at Scale
41mArchitecture Around Search
43mThe Search for a New Visual Search, Beyond Language
40mLevel Up: Everything you need to know about search, ...
26mStreamlining Property Search Across Multiple Platforms
26mOpenSearch and the Path to a Trillion-s...
41mScaling Search and Powering Vector-Based Retrieval With OpenSea...