Allocator-Aware (AA) Software
summary
Allocator-aware infrastructure yields large, measurable wins: local allocators cut global locking and cache diffusion, producing up to 4–5x speedups in examples and turning a 9s teardown into effectively zero; benchmarks trace back to ACCU’17 and internal tests. Tradeoffs include ~10% more source for BDE-style plumbing, training, tooling, larger interfaces, and misuse risks (monotonic pools can leak if mis-scoped). Presents concrete options: C++11 compile-time, C++17 PMR runtime-polymorphic, BDE wrapper style, and a language-based C++23 alternative under development.
AI-written summary · about
chapters (14)
- 0:00Introduction and book update
- 5:21Allocator basics and costs
- 8:52First-class vs economy software
- 19:49Economics analogy explained
- 24:48Allocator styles overview
- 25:09Performance benefits of allocators
- 31:28Real costs of allocator infrastructure
- 40:20Mitigating factors and automation
- 47:29Collateral benefits of allocators
- 59:49Concerns and historical context
- 1:07:59Zero-overhead principle debate
- 1:14:34Compatibility with modern C++
- 1:21:52Move semantics and allocators
- 1:25:04Conclusion and future outlook
related talks
01h 13mValue Proposition: Allocator-Aware (AA) Software
01h 01mGetting Allocators out of Our Way
01h 26mFrictionless Allocators
01h 41mLocal (arena) Memory Allocators
01h 01mAllocators: The Good Parts
57m“An allocator model for std2”
27m“From security to performance to GPU programming..."
34mAllocator-Aware C++ Type Design