rmalloc is designed to be a fast concurrent memory allocator with low fragmentation, low latency and high throughput. rmalloc draws inspiration from mainly mimalloc, jemalloc, slab allocator and tcmalloc. rmalloc was created by Romario Newell as an exploration into the memory allocator design space. Romario designed this allocator with simplicity in mind making it approachable for anyone wanting to understand and develop a viable production-quality memory allocator.
-
rmalloc is small and consistent. rmalloc is about 6k LOC making it easy to digest. Additionally, rmalloc code is well documented.
-
fast and highly concurrent. rmalloc outperforms jemalloc, tcmalloc and ptmalloc2 on xmalloc-testN benchmark. This is significant since xmalloc-testN benchmark is known to really stress test allocators.
-
smart recycling. Active empty slabs in rmalloc can be reused by different threads. Also threads participate in the cooperative recycling to return idle memory to the operating system. This keeps memory usage low.
No memory allocator is complete without a benchmark against mimalloc, jemalloc, tcmalloc or ptmalloc2. rmalloc was benchmarked against the aforementioned allocators. We provide a few benchmarks below were rmalloc is highly competitive or even better in some instances against established allocators.
git clone https://github.com/newell-romario/rmalloc.git
cd rmalloc
mkdir build && cd build
cmake -DCMAKE_BUILD_TYPE=RELEASE ..
make
After building, use directly from build directory.
LD_PRELOAD=./build/lib/librmalloc.so ./myprogram
