Speeding Up the Index in VK Video Search
summary
Describes accelerating the candidate-selection index for VK Video search to handle billions of documents. Explains engineering techniques and system-level optimizations used to make the retrieval layer faster and more efficient so the final ranking stage receives high-quality candidates.
AI-written summary · about
chapters (6)
- 0:00Introduction
- 3:23Search index structure
- 11:48Frequent query problem
- 17:51Max Block Skip optimization
- 26:43Results and new features
- 27:04Q&A
related talks
48mResearch Under the Microscope: How the Quality Lab Helps Improve VK Video
44mVK Video Delivery: From 10 Gbps over HTTP to 160 Gbps over HTTPS from a Single Machine
47mHow We Made VKontakte's Mobile Version Four Times Faster
14mUsing Performance Signals to Improve Video Recommendations
53mReindexer - a very fast in-memory DB with full-text search
18mDistVS: Large-scale Vector Search with Compute-Memory Disaggregation
34mMake It Under 100 Milliseconds: Contextual Ads on Sphinx
40mIn Search of... A better, fast