Local (Arena) Memory Allocators Part 1
summary
Claims that replacing a single global allocator with targeted local allocators can yield order-of-magnitude speed improvements, and measures that claim with benchmarks on allocator strategies (global, monotonic, multi‑pool, and combinations) and a C++17 polymorphic allocator interface. The talk introduces six dimensions to characterize subsystems (fragmentability, density, variation, utilization, locality, contention), shows empirical run‑time reductions across scenarios, and demonstrates a monotonic and multi-pool allocator design that actually speeds long‑running systems.
AI-written summary · about
chapters (9)
- 0:00Start
- 1:11Are memory allocators really worth the trouble? What situations merit their use?
- 8:31What are the common arguments against?
- 16:11(Main) Memory Segments
- 16:30Important Questions What is a Memory Allocator?
- 18:12A memory allocator organizes a' region of computer memory, dispensing and reclaiming authorized access to suitable sub-regions on demand. possibly non-contiguous
- 21:52A memory allocator is a stateful utility or mechanism that organizes a' region of computer memory, dispensing and reclaiming authorized access to suitable sub-regio on demand
- 25:49A memory allocator is (the client-facing interface for) a stateful utility or mechanism that organizes a' region of computer memory, dispensing
- 50:01What aspects of software affect optimal allocation strategy
related talks
01h 41mLocal (arena) Memory Allocators
49mMaking Allocators Work, Part I
27m“From security to performance to GPU programming..."
51mUnderstanding Allocator Impact on Runtime Performance in C++
01h 08mMaking Allocators Work, Part II
51mWriting Allocation Free Code in C#
39mMemory Efficient Java
42mSelf-Awareness in C++ Code