Xiaohongshu’s HELMSMAN moves billion-scale vector search to all-flash servers
小红书技术REDtech · wechat · 2026-07-23
Xiaohongshu’s engine team presented HELMSMAN at OSDI 2026, a full-flash ANNS system designed to cut the cost of large-scale vector search without giving up online SLA. The system targets search, recommendation, advertising, and other high-QPS pipelines where in-DRAM indexes are becoming too expensive.
HELMSMAN combines a clustering-based ANN design, an ANNS-specific storage stack built on SPDK, learned pruning to predict how many clusters to probe, and a GPU/CPU heterogeneous build pipeline. In production, it reportedly replaced a workload that previously needed about 35,000 CPU cores and 350 TB of DRAM with roughly 40 all-flash servers, cutting hardware cost by more than 90% while keeping millisecond latency and strong tail behavior. The paper also reports 2–16× throughput gains over prior DRAM-SSD ANNS systems and much faster rebuild times for billion-scale indexes.
Related event: Xiaohongshu's HELMSMAN: All-Flash Vector Retrieval(2 posts)→
More from Infra
- SGLang MegaKernel + PD Pushes the TCO-Latency Frontier on GB300, SemiAnalysis Benchmarks Show — AccBalanced · 2026-10-27
- Anthropic hits $16.3B H1 revenue as compute efficiency jumps 5.2x in 18 months — daniel_mac8 · 2026-10-27
- Tenstorrent runs Proteina protein model at high throughput, claims no HBM shortage — gajesh · 2026-10-27
- YC-backed Hopper launches voice-agent inference cloud: 80 ms TTFT at $0.30/1M tokens — ycombinator · 2026-10-27
- A100 rental prices still rising in 2026 per a16z data, Jevons paradox hits compute — VraserX · 2026-10-27
- Anthropic S-1 reveals $800B+ in off-books future compute commitments — GaryMarcus · 2026-10-27