Performance testing or system benchmarks for jemalloc

Referred from https://github.com/daanx/mimalloc-bench/blob/master/bench/alloc-test/allocator_tester.h

Configure with the following arguments when building vineyard:

cmake .. -DBUILD_VINEYARD_MALLOC=ON -DBUILD_VINEYARD_BENCHMARKS=ON

Then make the following targets:

make vineyard_benchmarks

The artifacts will be placed under the ${CMAKE_BINARY_DIR}/bin/ directory:

./bin/bench_allocator_system
./bin/bench_allocator_mimalloc
./bin/bench_allocator_vineyard

The benchmark artifacts accept an optional arguments to control the iterations count (default value is 100000000):

./bin/bench_allocator_system 1000
./bin/bench_allocator_mimalloc 1000
./bin/bench_allocator_vineyard 1000

There are two kinds of allocators that we used in vineyard: jemalloc and mimalloc. You could do the following steps to reproduce the benchmark result.

Run the vineyard server with 8G shared meory.

export vineyard_socket=[please create a socket file here]
./vineyardd -socket=$(vineyard_socket) -size=8G

Run the mimalloc benchmark test, and the argument is the iteration count.

# bench system
./bin/bench_allocator_system 1000
# bench mimalloc
./bin/bench_allocator_mimalloc 1000
# bench vineyard
./bin/bench_allocator_vineyard 10000

The next table is benchmark test result in the machine which has 8 physical cores and 64GB of memory.

nil means the allocator can't malloc the memory.