Hashmap Benchmark

Benchmarking multiple concurrent hashmap implementations.

Getting the source

This project uses multiple git submodules both for hashmap libraries and various utilities and they need to get initialized.

git clone https://github.com/dminikm/HashmapBenchmark.git
git submodule init
git submodule update

Building

mkdir build
cd build
cmake ..
cmake --build . --config Release --target install

If you are using mingw on windows (MSYS2/mingw-w64/...), you might need to configure turf to use it's WIN32 implementation to avoid build errors.

mkdir build
cd build
cmake -G "MinGW Makefiles" -DTURF_USERCONFIG=turf_options/mingw/turf_userconfig.h.in -DCMAKE_BUILD_TYPE=Release -DCMAKE_CXX_FLAGS=-m64 ..
cmake --build . --config Release --target install

Running

Wordcount test

Wordcount has 4 implementations:

• libcuckoo
• tbb-hash - tbb::concurrent_hash_map
• tbb-unordered - tbb::concurrent_unordered_map + tbb::atomic
• std-blocking - std::unordered_map + std::mutex

Example

Running libcuckoo benchmark with 4 threads, 60 runs outputting results to json:

mkdir runs
./HashmapBenchmark wordcount -t 4 -r 60 --implementation=libcuckoo --json=runs/4t_60r_libcuckoo.json --dataset=../data/test.ft.txt.out