MBW determines the read, write, and copy bandwidth available to single- and multi-threaded userspace programs, optionally taking NUMA placement into account. It can either utilize simple for loops and memcpy, or run (experimental!) AVX512 read / write / copy benchmarks.
This is an extended version of the benchmark originally developepd by Andras Horvath et al., available at http://github.com/raas/mbw / https://github.com/Willian-Zhang/mbw. Multi-threading, NUMA, and read/write support were not present in the original versions.
Mirrors of the MBW repository are available at the following locations.