USDB: Scaling Storage for Next-Generation Recommender Systems at ByteDance

Automated excerpt

Abstract:Recommendation models consume increasingly long, attribute-rich user histories, making small continuous updates and repeated large reads a central storage asymmetry. These designs reuse maintenance results across ingestion, inference, and training. Controlled comparisons with an Optimized RocksDB sequence baseline show faster long-history reads across projections and client loads.

Selected automatically from source text; not independently written or fact-checked. Read the original for full context.