Value Proposition: Allocator-Aware (AA) Software
summary
Argues that a company-wide allocator-aware software infrastructure (BDE, BT-verified tools, PMR polymorphic memory resource) delivers nearly first-class performance at a tiny fraction of per-team development cost, while enabling rapid prototyping, mmap placement and fine-grained instrumentation (test/counting allocators). Shows monotonic, pooled and managed allocators that preserve locality (which dominates allocation cost), cites internal benchmarks and a historical ~10% code bloat estimate for conversion, and proposes Bloomberg’s 2020 vision automation to eliminate most upfront plumbing.
AI-written summary · about
chapters (13)
- 0:00Intro
- 1:48Purpose of this Talk
- 12:37Style Alternatives
- 22:06Thread Locality
- 25:03Creating and Exploiting AA
- 27:53Up-Front (LIBRARY DEVELOPMENT) Costs
- 42:59Testing and Instrumentation
- 46:44Pluggable Customization
- 47:57Outline
- 50:48Why the Quotes?
- 53:51State-of-the-Art Global Allocators
- 54:47Zero-Overhead-Principle Compliance
- 59:46Verification/Testing Complexity
related talks
01h 31mAllocator-Aware (AA) Software
57m“An allocator model for std2”
01h 26mFrictionless Allocators
01h 01mAllocators: The Good Parts
01h 25mTaming Dynamic Memory - An Introduction to Custom Allocators
27m“From security to performance to GPU programming..."
49mMost Malleable Memory Management Method in C++
01h 03mMost Malleable Memory Management Method