New C++20 LRU Hash Table Optimizes AI Compute Infrastructure
TL;DR. A new C++20 LRU Hash Table offers zero runtime allocations and multi-core scalability for demanding computing tasks. - The table employs custom spinlocks and sharded architecture to enhance throughput on high-core CPUs. - Designed for HFT, real-time networking, and kernel components, it avoids standard library bottlenecks. - NUMA-aware memory distribution and lazy LRU promotion further boost performance.
- High-performance concurrent LRU hash table developed for demanding system workloads.
- Features zero runtime allocations, custom TTAS spinlocks, and sharded architecture for multi-core scalability.
- Addresses standard library bottlenecks like global lock contention and pointer chasing.
- Supports dual environments including Windows Kernel mode, targeting HFT, network infrastructure, and kernel components.
Sources
- High-Performance Array-Backed LRU Hash Table — github.com