On Aug 03, 2021 / 15:58, Ankit Kumar wrote:
> Existing random zone index generation logic is dependent on the file size.
> For smaller I/O sizes the random zone index always return a particular
> section of open zones. As this index is used to return one of the open zones,
> it was observed that after one of the max_open_zones / job_max_open_zones limit
> is reached all further I/O's are redirected to only a few open zones till they
> are full.
>
> This patch modifies the random zone index genration logic so that it is uniform
> across all the open zones.
> ---
> zbd.c | 5 +++--
> 1 file changed, 3 insertions(+), 2 deletions(-)
>
> diff --git a/zbd.c b/zbd.c
> index 04c68de..43f12b4 100644
> --- a/zbd.c
> +++ b/zbd.c
> @@ -1184,11 +1184,12 @@ out:
> return res;
> }
>
> -/* Anything goes as long as it is not a constant. */
> +/* Return random zone index for one of the open zones. */
> static uint32_t pick_random_zone_idx(const struct fio_file *f,
> const struct io_u *io_u)
> {
> - return io_u->offset * f->zbd_info->num_open_zones / f->real_file_size;
> + return (io_u->offset - f->file_offset) * f->zbd_info->num_open_zones /
> + f->io_size;
> }
One nit I suggest is to remove the period in the commit message title. Other
than that, I confirmed the change is good. I confirmed that it improves the
randomness of the open zone choice. It reverts the part of the commit 6463db6c1
('fio: fix interaction between offset/size limited threads and
"max_open_zones"'), but I confirmed that still the intent of the commit
6463db6c1 is kept after applying this change.
Tested-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@xxxxxxx>
--
Best Regards,
Shin'ichiro Kawasaki
