Reducing the p99 latency of a vector index
Why the tail degrades before the median: efSearch, metadata filtering, reindexing, and how memory pressure shows up in high percentiles.
ET
Author
Engineering team
Vela Engineering
The collective that designs and operates Vela's indexing engine and API
About this team
The engineering team brings together the people who build and run Vela: the vector storage layer, HNSW graph construction and maintenance, the query planner, metadata filtering, and cluster operations in production.
Articles signed by this team are collective engineering notes. They describe implementation choices, deliberate trade-offs and incidents we learned from. None of these pieces are attributed to any one person.
Published