Xiaohongshu's RedDB: unified multimodal DB with >50% throughput gain and 50% cost cut
小红书技术REDtech · wechat · 2026-11-17
Xiaohongshu's storage team details RedDB, a distributed multimodal database unifying wide-table, vector and graph models on one foundation: a 4-layer architecture with Raft replication, sharding, elastic scaling, RocksDB-based row engine, backpressure flow control, and sparse-scan optimizations. Benchmarks show median throughput gains over 50%, 60% P99 latency reduction, and 50%+ cluster cost savings. Future plans include unified metadata, index-change-stream maintenance, query coordination, storage-compute separation, and cross-modal compute.
More from Infra
- How rollout staleness hurts on-policy distillation, and how to build a faster pipeline — burkov · 2026-11-17
- SK hynix's Nvidia revenue doubled to $24B, but US customer concentration is the real risk — tengyanAI · 2026-11-17
- macOS TCP Stack Breaks After Exactly 49.7 Days of Uptime Due to 32-bit Timestamp Overflow — DanielLockyer · 2026-11-17
- Running a 2M-param embedding model on an ESP32 with no_std Rust at 830 embeddings/s — qdrant_engine · 2026-11-17
- Voice agent failed 0/5 turns in background noise; adaptive client-side VAD gate fixed it at lower latency — shrapknife · 2026-11-17
- Strix Halo int4×int4 WMMA: Gemma 4 31B at 2.43× llama.cpp with matched quality — No-Dragonfly6246 · 2026-11-17