Benchmarks of NetBSD 10 show huge performance and scalability gains over NetBSD 9.x, especially on multiprocessor and multicore systems, for compute and filesystem-bound applications. Areas of improvement included: switched the kernel's file path lookup cache to use faster per-directory red-black trees; improved scheduler performance, including the ability to more appropriately spread load on a mixture of slow and fast cores (e.g. big.LITTLE Arm CPUs); various optimizations for the machine-independent virtual memory system; improved tracking of clean/dirty pages, speeding up fsync(2) on large files by orders of magnitude; improved parallelization - rewritten page allocator with awareness of CPU topology, replaced global counters with per-CPU counters, and reduced lock contention; improved the performance of the select(2) and poll(2) system calls; improved the performance of tmpfs; implemented lazy update of atime/mtime; various optimizations of architecture-dependent x86 and AArch64 code, vastly improved network and I/O throughput on aarch64....
Official announcement
Download NetBSD
