Sitsofe Wheeler

Thanks for checking @louwrentius. Thinking more I believe the issue is down to averaging. The following question illustrates this: "If five 200ms periods report their 'IOPS' as 10, what are...

It's probably too late to change the units for steady state bw (otherwise it breaks all existing users) but I would guess another key could be introduced called bw_kb that...

@louwrentius Well I guess my thinking was that if we introduce `bw_kb` we could then have a transition period and then remove the old value entirely at some point...

@louwrentius so digging around I've found https://github.com/axboe/fio/issues/378#issuecomment-306223858 . Looking closer, there is a mishmash of bytes/kbytes scattered across the JSON output which is defeating consistency... What are your thoughts on...

@Gt3pccb [It doesn't make sense to use direct=1 and buffered=1](https://fio.readthedocs.io/en/latest/fio_man.html#cmdoption-arg-buffered) in the same job - they are antonyms of each other...

@Gt3pccb direct and buffered are antonyms in **fio** and [map to the same option (direct)](https://github.com/axboe/fio/blob/738694ae75722aee2aa3952257af62aee7957770/options.c#L2789-L2795). If you want to control writethrough I believe that's done with `sync=1` with the windowaio...

@Gt3pccb I'm not sure I follow. For the four cases you defined with the options ([direct](https://fio.readthedocs.io/en/latest/fio_doc.html#cmdoption-arg-direct), [sync](https://fio.readthedocs.io/en/latest/fio_doc.html#cmdoption-arg-sync)) that fio has already when using the windowsaio engine: 1. Buffered + write-through...

@sathish356 The code backs up what you're saying... If you look at - https://github.com/axboe/fio/blob/fio-3.27/backend.c#L2146-L2153 - https://github.com/axboe/fio/blob/fio-3.27/server.c#L959-L978 you will see that fio tells running threads to terminate after a trigger goes...

@garyguo We're not a minio users but are you happy to debug this one. Does moving stonewall to the global section make a difference?

Mentioning @l-mb for comment.