Re: Help required to understand fio output

[Jens Axboe <jens.axboe@oracle.com>]

On Wed, Nov 19 2008, Erwan Velu wrote:
> Jens Axboe wrote:
> 
> [...]
> >The output could do with a bit of prettying up... It's explained in the
> >documentation, though, The 100=64.24% field means that 64.24% of the ios
> >completed within 50 and 99ms, as you indicate.
> >  
> I would say that [50-99]=64.24% would have been less confusing even if 
> it take more room.
> [...]

That's the primary problem, but I guess it's just to compressed for its
own good...

> >>And then, how should I understan lines 8 & 9 that reports that 100% of 
> >>IOs were completed with a depth=4... What happened ?
> >>    
> >
> >Same deal, 4 means 1-4 completions per io_getevents() call for your test
> >case. So any attempt to retrieve completed IO from the system resulted
> >in a return of 1-4 ios.
> I just don't understand what the reason of that 64 -> 4 depth reduction. 
> Is that related to the storage or the kernel capabilities ?

If you don't tell fio otherwise, it'll attempt to reclaim completed IO
as soon as it can. You can control this behaviour with the iodepth_*
options.

-- 
Jens Axboe