yuvalif / memory-benchmark

Memory management benchmark for a specific workload

Memory Benchmarking

This allows for benchmarking of: libc, tcmalloc and jemalloc.

• benchmarking is against a multi-threaded workload, where memeory is allocated in one thread and released in another
• different object sizes may be used to create a more realistic memory management scenario
• number of threads is configurable
• no mutexes are used for thread synchronization, to minimize contention overhead

Prerequisites

• cmake is used
• tcmalloc and jemalloc needs to be installed on the machine. E.g.

sudo dnf install jemalloc-devel
sudo dnf install google-perftools
sudo dnf install numactl

• argument parsing is done using Argh, which is cloned as a submodule, so don't forget to update it before building:

git submodule update --init

Building:

To build, create a build directory. E.g.

mkdir build

And run cmake and make there. E.g.

cd build && cmake .. && make -j4

Running:

Three binaries are generated: proxy_libc, proxy_tc, proxy_je. The all share the same options:

$ ./proxy_libc -h
usage: ./proxy_libc[--iterations=<number>|--threads=<number>|--min-size=<bytes>|--max-size=<bytes>|--overall-iterations=<number>]

	iterations: the number of objects each producer is going to create [default 1M]
	threads: the number of consumer threads and number of producer threads. zero means a single thread [default 0]
	min-size: the minimum size in bytes to allocate per object [default to max-size if set, or to 1K if not]
	max-size: the maximum size in bytes to allocate per object [default to min-size if set, or to 1K if not]
	overall-iterations: number of times the test is executed [default 1]

To get a comperative report of operations per second for different setups, you can use the run_test.sh script. Notre that script should be executed from within the build directory. E.g.

cd build && ../run_tests.sh