Re: fio replay

[Sitsofe Wheeler <sitsofe@yahoo.com>]

Hi,

I've been using blktrace with fio by doing the following on Fedora 20
with a git version of fio (fio-2.1.5-16-g81fa):
blktrace -d /dev/sdd
blkparse sdd -O -d sdd.dump
fio --read_iolog=sdd.dump --ioengine=libaio --name=replay

Using iowatcher I can see that fio is hitting the same blocks that the
blktrace recorded.

It is also worth noting that certain options like rate, rate_iops and bs
do not seem to affect replay jobs. Additionally, fio does not seem able
to replay a job with the same timings that it is was recorded with (it
also seems to do so faster) whereas btreplay has no issue doing this
(even reproducing similar depths to the original job).

The one minor issue that I see is this:

Run status group 0 (all jobs):
   READ: io=40549MB, aggrb=30919KB/s, minb=30919KB/s, maxb=30919KB/s, mint=1342918msec, maxt=1342918msec
  WRITE: io=40562MB, aggrb=30929KB/s, minb=30929KB/s, maxb=30929KB/s, mint=1342918msec, maxt=1342918msec
fio: file hash not empty on exit

Why is the file hash not empty on exit? This happens no matter what I
do.

On Wed, Feb 26, 2014 at 02:23:27PM +0000, Lamkin, Eric wrote:
> 
> Thanks for responding.
> I sent the below out last week but didn't get a reply so thought I'd
> try a simpler 'what works' tactic this morning.  As seen in the below
> I'm using 2.0.10.
> 
> > On Feb 26, 2014, at 7:44 AM, "Lamkin, Eric" <Eric.Lamkin@lsi.com> wrote:
> > 
> > Is there a version of fio where replay works with a Linux blktrace?
> > 
> anything within the last few years should play them back just fine, is
> it broken in a specific version you're trying?

-- 
Sitsofe | http://sucs.org/~sits/