* [PATCH 1/2] eta: Avoid adjustements to a negative value
@ 2018-09-26 6:42 Damien Le Moal
2018-09-26 14:26 ` Jens Axboe
0 siblings, 1 reply; 2+ messages in thread
From: Damien Le Moal @ 2018-09-26 6:42 UTC (permalink / raw)
To: fio, Jens Axboe
In the case of zonemode=strided job with both zone_size and zone_skip
specified, thread_eta() calculates the upper bound of the number of
zones that will be processed, including skipped bytes between zones
(zone_skip). Adjusting bytes_total (i.e. total_io_size) by substracting
this number of zones times the amount of skipped bytes can result in
a negative value for bytes_total (a large number of bytes) when for
example the job operates on an entire disk with a capacity that is not
divisible exactly by zone_size+zone_skip or if the options --size is
used to limit the I/O range. In such case, use the lower bound of the
number of zones to obtain a better approximation of the job eta.
Additionnally, if --io_size was specified, bytes_total will indicate
this exact value, so adjusting that value for zonemode != none is not
necessary.
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
---
eta.c | 21 ++++++++++++++++++---
1 file changed, 18 insertions(+), 3 deletions(-)
diff --git a/eta.c b/eta.c
index 970a67df..b69dd194 100644
--- a/eta.c
+++ b/eta.c
@@ -177,12 +177,27 @@ static unsigned long thread_eta(struct thread_data *td)
bytes_total = td->fill_device_size;
}
- if (td->o.zone_size && td->o.zone_skip && bytes_total) {
+ /*
+ * If io_size is set, bytes_total is an exact value that does not need
+ * adjustment.
+ */
+ if (td->o.zone_size && td->o.zone_skip && bytes_total &&
+ !fio_option_is_set(&td->o, io_size)) {
unsigned int nr_zones;
uint64_t zone_bytes;
- zone_bytes = bytes_total + td->o.zone_size + td->o.zone_skip;
- nr_zones = (zone_bytes - 1) / (td->o.zone_size + td->o.zone_skip);
+ /*
+ * Calculate the upper bound of the number of zones that will
+ * be processed, including skipped bytes between zones. If this
+ * is larger than total_io_size (e.g. when --io_size or --size
+ * specify a small value), use the lower bound to avoid
+ * adjustments to a negative value that would result in a very
+ * large bytes_total and an incorrect eta.
+ */
+ zone_bytes = td->o.zone_size + td->o.zone_skip;
+ nr_zones = (bytes_total + zone_bytes - 1) / zone_bytes;
+ if (bytes_total < nr_zones * td->o.zone_skip)
+ nr_zones = bytes_total / zone_bytes;
bytes_total -= nr_zones * td->o.zone_skip;
}
--
2.17.1
^ permalink raw reply related [flat|nested] 2+ messages in thread
* Re: [PATCH 1/2] eta: Avoid adjustements to a negative value
2018-09-26 6:42 [PATCH 1/2] eta: Avoid adjustements to a negative value Damien Le Moal
@ 2018-09-26 14:26 ` Jens Axboe
0 siblings, 0 replies; 2+ messages in thread
From: Jens Axboe @ 2018-09-26 14:26 UTC (permalink / raw)
To: Damien Le Moal, fio
On 9/26/18 12:42 AM, Damien Le Moal wrote:
> In the case of zonemode=strided job with both zone_size and zone_skip
> specified, thread_eta() calculates the upper bound of the number of
> zones that will be processed, including skipped bytes between zones
> (zone_skip). Adjusting bytes_total (i.e. total_io_size) by substracting
> this number of zones times the amount of skipped bytes can result in
> a negative value for bytes_total (a large number of bytes) when for
> example the job operates on an entire disk with a capacity that is not
> divisible exactly by zone_size+zone_skip or if the options --size is
> used to limit the I/O range. In such case, use the lower bound of the
> number of zones to obtain a better approximation of the job eta.
>
> Additionnally, if --io_size was specified, bytes_total will indicate
> this exact value, so adjusting that value for zonemode != none is not
> necessary.
Applied 1-2, thanks Damien.
--
Jens Axboe
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2018-09-26 14:26 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2018-09-26 6:42 [PATCH 1/2] eta: Avoid adjustements to a negative value Damien Le Moal
2018-09-26 14:26 ` Jens Axboe
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.