Re: skip file setup?

Looks like we don't call posix_fallocate if new_layout == 0, which is true
for writes with overwrite == 0. We truncate (takes no time), but no
posix_fallocate or extend. This works for me.

On Nov 26, 2007 11:37 PM, Jens Axboe <jens.axboe_at_oracle.com> wrote:

> On Mon, Nov 26 2007, Shawn Lewis wrote:
> > Nevermind, looks like write phases don't actually extend the file, which
> is
> > what I'm looking for. :) Maybe there was an older version of fio that
> didn't
> > do this, or I'm just crazy.
>
> Only option in that area is 'overwrite', which controls whether the
> write phase is an overwrite or new write. The file setup will still
> ftruncate() and posix_fallocate() the blocks. How about an option for
> skipping the fallocate? It'll definitely result in a zero fill for lots
> of systems.
>
> >
> > Thanks!
> > Shawn
> >
> > On Nov 26, 2007 8:14 PM, Shawn Lewis <shawnlewis_at_google.com> wrote:
> >
> > > Jens,
> > >
> > > Does any combination of fio options allow us to completely skip file
> > > setup? I'd like to have a thread with rw=write size=300g that doesn't
> ever
> > > do file setup. This is useful if you want to have a write phase and
> then a
> > > verify phase and you don't care about performance. If this doesn't
> exist
> > > I'll send a patch.
> > >
> > > It's also useful because runtime can't cap the time for the file setup
> > > phase, so if you have a really slow disk it'll take forever to lay the
> file
> > > out.
> > >
> > > Thanks,
> > > Shawn
> > >
>
> --
> Jens Axboe
>
>
Received on Wed Nov 28 2007 - 02:39:47 CET