How flexible I/O tester uses measurement units
----------------------------------------------

fio uses the opposite symbol for kibibytes/kilobytes (Kb/KiB) than
ISO 80000-1 by default in *jobfiles*. The user readable output however
adheres to the standard, while the parseable output for scripts also
uses the opposite symbol.

This leads to weird scenarios like:

% fio --name=rand-read --bs=4k --size=1MiB --iodepth=64 --runtime=10 --rw=randread
rand-read: (g=0): rw=randread, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=psync, iodepth=64
[…]
% ls -l rand-read.0.0                                                             
-rw-r--r-- 1 root root 1000000 Mär 26 13:10 rand-read.0.0

In case you decide to write fio jobs in a way that adheres to the ISO
standard, you can use:

% fio --name=rand-read --kb_base=1000 --bs=4KiB --size=1MiB --iodepth=64 --runtime=10 --rw=randread
rand-read: (g=0): rw=randread, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=psync, iodepth=64
[…]
% ls -l rand-read.0.0                                                                              
-rw-r--r-- 1 root root 1048576 Mär 26 13:19 rand-read.0.0

Note: In this case the block size needs to be in "KiB" as well, otherwise
it would be just 4000 bytes in this example.

I received no feedback from upstream developers regarding this issue. As I
decided to keep upstream behavior rather than diverting from it, using
'--kib_base=1000' is your option in case you decide to adhere to the
standard. For further discussion see:

fio: uses the opposite symbol for kibibytes/kilobytes (Kb/KiB) than ISO 80000-1
https://bugs.debian.org/872321

-- Martin Steigerwald <martin.steigerwald@proact.de>, Mon, Mar 26 13:29:45 +0200