Turbopuffer Abandons Vector-First Architecture for Better Performance
Turbopuffer, a serverless vector database, is overhauling its storage architecture to improve search performance across text, regex, and vector queries. The company is moving away from its vector-primary index design, which has constrained certain query plans like GROUP BY and aggregations. The new architecture will treat the Approximate Nearest Neighbor (ANN) index as a secondary index, allowing for better storage and write amplification, and enabling larger block sizes for vectorized query processing. This shift is expected to significantly enhance performance for non-vector query types. Turbopuffer has already achieved 100% CI passes on its new v3 architecture, marking a major milestone in this transition.
Read the full article
Continue reading at Hacker News →This is an AI-generated summary. Read the original for the full story.