Tailslayer: open-source C++ library kills DRAM refresh stalls, cutting p99.9 latency up to 18.5x

thisdudelikesAI · x · 2026-10-29

A developer has open-sourced Tailslayer (Apache 2.0, 2.9k GitHub stars), a C++ library that eliminates latency spikes caused by DRAM refresh stalls.

• The problem: DRAM refreshes its charge thousands of times per second, and any read landing in that window stalls while the CPU sits frozen — the source of unexplainable p99/p999 latency spikes.
• The fix: using undocumented DRAM channel scrambling offsets, Tailslayer replicates data across independent channels with uncorrelated refresh schedules, then fires hedged reads across all replicas; whichever responds first wins, making stalls irrelevant.
• Usage: works on AMD, Intel, and Graviton; the API takes just a signal function and a work function — the library handles replication, core pinning, and hedged reads. A bundled trefiprobe.c lets you watch stalls appear and vanish in real time.
• Benchmarks (see chart): N-way hedged reads on EPYC 9255/9R45, Xeon 8480C, and Graviton 3/4 cut p99.9 latency by up to 18.5x (Graviton V2, 12 channels), with diminishing returns per platform.

Useful for HFT, real-time systems, and game servers — anywhere tail latency matters.

Original post →

More from Infra

• AI Infrastructure Bond Yields Hit Double Digits as Credit Markets Near Limits — GaryMarcus · 2026-10-29
• Communication Is the Bane of Inference and Training: One Beefy GPU Beats Multi-Node — TheZachMueller · 2026-10-29
• Analyst: skip Qualcomm Dragonwing—it's branding, not a real platform — DavidLinthicum · 2026-10-29
• Europe's data center boom sparks a fight over who gets grid power first — kindermaxi123 · 2026-10-29
• Reddit asks: what have you actually built with local LLMs? — Loose_Doubt367 · 2026-10-29
• Minigraf: SQLite-style single-file bi-temporal graph database for AI agent memory — JeremyCMorgan · 2026-10-29

Infra channel →