Kuaishou System Software Team: Architecture, Innovations, and Performance Optimizations
This article presents an overview of Kuaishou's system software team, detailing its core value, challenges, the "Four Horizontal Four Vertical" architecture, JVM innovations such as transparent coroutines and checkpointing, advanced compilation techniques, kernel isolation for cloud‑native workloads, and extensive performance‑boosting strategies that have dramatically improved resource utilization and stability.
System software serves as the foundation of software architecture, bridging hardware and higher‑level software; Kuaishou's system software team has spent four years building innovations across Java coroutines, compiler optimizations, and OS improvements.
Core Value and Challenges – System software enhances server resource utilization, improves performance by 30‑50%, and ensures business stability, but scaling introduces stricter constraints, requiring deep low‑level expertise and precise problem targeting.
"Four Horizontal Four Vertical" Architecture – The team established a layered tech stack: JVM runtime, C++ build system, OS layer, and micro‑architecture layer, each critical for stability and performance.
JVM Innovations – Developed Java 17 transparent coroutines (non‑intrusive, 30% throughput gain) and a JVM checkpoint feature that cuts startup time from 150 s to 80 s and reduces launch CPU utilization from 23% to 5%.
Compilation Techniques – Integrated BOLT, AutoFDO, LTO, and other optimizations, achieving 10‑13% overall performance gains and 4‑8% latency reductions; combined BOLT + AutoFDO with a Match+Infer solution raised optimization impact from 8% to 12%.
Kernel Isolation & Resource Scheduling – Built a custom kernel for cloud‑native scenarios, introducing online/offline scheduling trees, dual‑queue QoS, and LLC isolation, which doubled machine utilization while protecting online services.
Extreme Performance Optimizations – Applied CPU‑level IPC, utilization, and complexity reductions, refined a critical size() function to cut its CPU share from 8.4% to 2.5% and doubled overall performance, saving thousands of servers.
In summary, the team’s breakthroughs in system software have delivered substantial cost savings, performance improvements, and a robust foundation for Kuaishou’s large‑scale services, while continuing to invite talent to further advance these technologies.
Kuaishou Tech
Official Kuaishou tech account, providing real-time updates on the latest Kuaishou technology practices.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.