What type of data does IOzone test with? Random? Zeros?

[scurrier]

I am doing file system testing on a ZFS dataset. I left the default LZ4 compression enabled without thinking about it. Now that I just thought about it and I suspect my long-running IOzone test may be worthless because compression is currently showing to be 32x. So I'm guessing IOzone does not use random data? There's no mention of what type of data is used in the IOzone manual.

For reference, here's the command I used:

iozone -g 100G -Rab iozone_noshare.wks

 

[eraser]

Good catch. You'll want to add the following iozone command line parameters to use incompressible data: "-+w 1 -+y 1 -+C 1"

Here is the full command line I like to use to quickly test (change the -s 500 M to a larger size if you want a larger test): "-+w 1 -+y 1 -+C 1 -+n -e -c -t 1 -r 512K -s 500M -i 0 -i 2"

Actually the FreeNAS documentation should probably be updated to suggest using those parameters now that it enables compression by default. I'll open a bug/feature request later on to do that.

edit: turns out that the above suggested command line parameters don't work as expected and still generate highly compressible test data. Don't use the above suggested parameters...

 

[eraser]

Looking back, I don't know why I went with a 1% compressible data. Suggest changing the "-+w 1 -+y 1 -+C 1" section to "-+w 0 -+y 0 -+C 0".

Opened following feature request to update documentation: https://bugs.freenas.org/issues/4619

edit: turns out that the above suggested command line parameters don't work as expected and still generate highly compressible test data. Don't use the above suggested parameters...

 

[viniciusferrao]

Damn, this explains a lot!

 

[eraser]

Looking back, I don't know why I went with a 1% compressible data. Suggest changing the "-+w 1 -+y 1 -+C 1" section to "-+w 0 -+y 0 -+C 0".

Opened following feature request to update documentation: https://bugs.freenas.org/issues/4619

Oops.... Using the above suggested "0" seems to result in very highly compressible data. And, after checking, using "1" also results in data that compresses much better then expected. Not sure why this is, but until it is figured out I guess the best thing would be to (temporarily) disable compression on your ZFS dataset before testing and/or find another testing tool that properly generates random data.

I have updated the wiki to reflect this information and also updated https://bugs.freenas.org/issues/4619 . Sorry about the mistaken suggestion.