m4drat / memplusplus-benchmarks

Set of benchmarks for mem++ allocator

memplusplus-benchmarks

What is it

This repo provides a set of different benchmarks for memplusplus memory allocator. Each test benchmarks a wide range of functionality starting from simple allocation/deallocation speed and ending with some complex things (such as speed of data access after heap compacting/relayouting).

How to run

1. Clone this repo: git clone https://github.com/m4drat/memplusplus-benchmarks --recurse-submodules
2. Run all benchmarks cd memplusplus-benchmarks && ./compile_all_and_run.sh.sh

Benchmarks description and results

1. benchmark_alloc.cpp - Sequence of allocations from the same size bucket

   Time spent to allocate 4096 objects of random size in us:

   

2. benchmark_alloc_dealloc.cpp - Allocations and immediate deallocations

3. benchmark_dealloc.cpp - Sequence of chunks deallocations from the same size bucket

   Time spent to deallocate 4096 objects of random size in us:

   

4. benchmark_complex.cpp - Emulates a complex workload with allocations, deallocations and data access. This test is the most complex one and it is the most representative of real-world workloads. Benchmark inspired by rpmalloc-benchmark

   Time spent to perform 1m operations (approx. 0.5m allocations, 0.5m deallocations) in ms:

   

   Peak memory usage

   

   Time spent to perform 2m operations (approx. 1m allocations, 1m deallocations) in ms:

   

   Peak memory usage

   

5. benchmark_memory_access.cpp - Data access speed before/after compacting (only for memplusplus)

   First column - optimally layouted and accessed linked list
   second column - randomized linked list, but after layouting
   third column - randomized linked list

   Time spent to iterate over linked list of size 8192 in us:

   

   Time spent to iterate over linked list of size 16384 in us:

   

Targets

• gcpp
• jemalloc
• mempp
• mimalloc
• [?] ptmalloc2 - latest (glibc 2.36), currently uses libc 2.31 (from my machine)
• ptmalloc3
• rpmalloc