All of lore.kernel.org
 help / color / mirror / Atom feed
* Sequential write problems
@ 2012-04-01 19:31 Hoppetauet
  2012-04-02  4:41 ` Jens Axboe
  0 siblings, 1 reply; 2+ messages in thread
From: Hoppetauet @ 2012-04-01 19:31 UTC (permalink / raw)
  To: fio

Hello

I'm running some benchmarks on virtual machines
I made a script that runs fio N times, with the following job file

[seqwrite]
rw=write
size=${SIZE}
directory=${DIRECTORY}
bs=${BS}
overwrite=1
refill_buffers

The first run gives about 30MB/s, which is what dd tells me is correct 
for the disk at hand
however, from the second to last runs, I get about double that, which 
suggests some sort of caching effect

Is the data that's written to the file not random? I thought 
refill-buffers and overwrite would ensure that

The virtual machines are ubuntu oneiric 11.10, and my fio version is 2.06

Thanks


^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: Sequential write problems
  2012-04-01 19:31 Sequential write problems Hoppetauet
@ 2012-04-02  4:41 ` Jens Axboe
  0 siblings, 0 replies; 2+ messages in thread
From: Jens Axboe @ 2012-04-02  4:41 UTC (permalink / raw)
  To: Hoppetauet; +Cc: fio

On 2012-04-01 12:31, Hoppetauet wrote:
> Hello
> 
> I'm running some benchmarks on virtual machines
> I made a script that runs fio N times, with the following job file
> 
> [seqwrite]
> rw=write
> size=${SIZE}
> directory=${DIRECTORY}
> bs=${BS}
> overwrite=1
> refill_buffers
> 
> The first run gives about 30MB/s, which is what dd tells me is correct 
> for the disk at hand
> however, from the second to last runs, I get about double that, which 
> suggests some sort of caching effect
> 
> Is the data that's written to the file not random? I thought 
> refill-buffers and overwrite would ensure that

It is completely random data, and it's reseeded for every run. So with
the above job, there shouldn't be any chance to de-dupe or compress
anything. Maybe it's the layout? Fio defaults to using the same sequence
of random offsets everytime, to make a given run repeatable. You can set
randrepeat=0 to turn that off. That'll cause fio to random seedly the IO
offset generator as well, making the written patterns different from run
to run as well.

-- 
Jens Axboe


^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2012-04-02  4:41 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2012-04-01 19:31 Sequential write problems Hoppetauet
2012-04-02  4:41 ` 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.