ScaleMM

Uses thread-local storage to give each thread its own memory, drastically reducing contention and improving performance in benchmarks against FastMM and TopMM, as shown in scaling tests up to multiple cores.

Implements techniques like locality, reduced branch misprediction, compact code, and inlining, optimized with Intel VTune, leading to high-speed allocation and deallocation for small to medium blocks.

Organizes memory into fixed-size mini blocks (32-256 bytes) and small blocks (256-2048 bytes) for fast reuse with low overhead, using fixed arrays instead of linked lists to minimize cache misses.

Demonstrates clear scalability advantages in provided benchmarks, outperforming alternatives like FastMM and TCmalloc in multithreaded scenarios, with detailed charts showing improved throughput.