sysbench file IO benchmark

[SchumannFrequency]

I did the following test on my system via sysbench:

sysbench --test=fileio --file-total-size=30G prepare
sysbench --test=fileio --file-total-size=30G --file-test-mode=rndrw --init-rng=on --max-time=300 --max-requests=0 run

This is the result I see.
Total transferred 35.876Gb (124.81Mb/sec)

I only have 16 GB of RAM so I think a 30 GB test file is enough to eliminate RAM caching. (I'm not sure to be honest)
If you have more RAM you will probably need to use a much larger test file.

Hardware: Intel 12600KF (stock) -- Kingston 6200 MHz CL36 -- Sapphire RX 7600 -- BIOSTAR B760MZ-E PRO -- Antec P6 -- Xilence XP550 -- ARCTIC i35 -- 980 PRO 500GB
Software: OpenBSD, bspwm, open-source GPU driver, UFS file system (a.k.a. FFS)

As of sysbench 1.0 support for native Windows builds was dropped. It may be re-introduced in later releases. Currently, the recommended way to obtain sysbench on Windows is using Windows Subsystem for Linux available in Windows 10.

After installing WSL and getting into he bash prompt on Windows following Debian/Ubuntu installation instructions is sufficient. Alternatively, one can use WSL to build and install sysbench from source, or use an older sysbench release to build a native binary.

On BSD and Linux systems it is easy to use sysbench so I will not explain this in depth.

 

[SchumannFrequency]

sysbench --test=fileio --file-total-size=30G prepare
sysbench --test=fileio --file-total-size=30G --file-test-mode=rndrw --max-time=300 --max-requests=0 run

Result: read 10.17 MB/s, written 6.78 MB/s

I achieved the above result on the same hardware, but using the Samsung EVO 850 500GB (SATA) instead of the 980 PRO (NVMe).
The software is different, for this last result I use Calculate Linux + XFS and my first result (124.81Mb/sec) was on OpenBSD + UFS.