Inconsency between cli output and gnuplot

Inconsency between cli output and gnuplot

[Erwan Velu]

My 1.23 fio ouput display :

  read : io=32,003MiB, bw=131MiB/s, iops=1,362, runt=250326msec
    slat (usec): min=13, max=145, avg=16.69, stdev= 3.39
    clat (msec): min=31, max=129, avg=46.94, stdev= 7.24

clat output of gnuplot reports units in msec. The clat gnuplot output 
show a msec label on the y axis but trace some usec.

slat cli reports values that doesn't match at all what is viewable from 
gnuplot output. Averagage is said to be 16 on the cli whereas the 
gnuplot trace shows 20 at min.
Max is said to be 145 for slat but displayed at 60 on the gnuplot.
Slat is also said to be usec on the cli but shown as msec on the y axis 
of the gnuplot.

Complete archive showing that could be downloaded at 
http://konilope.linuxeries.org/tmp/localr.tar.bz2

It seems that units may vary from the cli output regarding the range of 
results. I have some case where clat is shown in usec instead of msec. 
Maybe that what confuse the gnuplot output.
Like in :
  write: io=32,003MiB, bw=124MiB/s, iops=1,294, runt=263681msec
    slat (usec): min=12, max=71, avg=15.43, stdev= 2.72
    clat (usec): min=663, max=1,537K, avg=49431.63, stdev=108787.82
    bw (KiB/s) : min=    0, max=171311, per=99.91%, avg=127152.93, 
stdev=21619.46

-- 
Erwan Velu
Pre-Sales Engineer
Seanodes
http://www.seanodes.com
+33 (0)1 41 22 13 83

Re: Inconsency between cli output and gnuplot

[Erwan Velu]

Erwan Velu wrote:
> My 1.23 fio ouput display :
>
>  read : io=32,003MiB, bw=131MiB/s, iops=1,362, runt=250326msec
>    slat (usec): min=13, max=145, avg=16.69, stdev= 3.39
>    clat (msec): min=31, max=129, avg=46.94, stdev= 7.24
>
> clat output of gnuplot reports units in msec. The clat gnuplot output 
> show a msec label on the y axis but trace some usec.
>
> slat cli reports values that doesn't match at all what is viewable 
> from gnuplot output. Averagage is said to be 16 on the cli whereas the 
> gnuplot trace shows 20 at min.
> Max is said to be 145 for slat but displayed at 60 on the gnuplot.
> Slat is also said to be usec on the cli but shown as msec on the y 
> axis of the gnuplot.
>
> Complete archive showing that could be downloaded at 
> http://konilope.linuxeries.org/tmp/localr.tar.bz2
>
> It seems that units may vary from the cli output regarding the range 
> of results. I have some case where clat is shown in usec instead of 
> msec. Maybe that what confuse the gnuplot output.
> Like in :
>  write: io=32,003MiB, bw=124MiB/s, iops=1,294, runt=263681msec
>    slat (usec): min=12, max=71, avg=15.43, stdev= 2.72
>    clat (usec): min=663, max=1,537K, avg=49431.63, stdev=108787.82
>    bw (KiB/s) : min=    0, max=171311, per=99.91%, avg=127152.93, 
> stdev=21619.46
>
What's your opinion on that ?

-- 
Erwan Velu
Pre-Sales Engineer
Seanodes
http://www.seanodes.com
+33 (0)1 41 22 13 83

Re: Inconsency between cli output and gnuplot

[Jens Axboe]

On Wed, Nov 19 2008, Erwan Velu wrote:
> My 1.23 fio ouput display :
> 
>  read : io=32,003MiB, bw=131MiB/s, iops=1,362, runt=250326msec
>    slat (usec): min=13, max=145, avg=16.69, stdev= 3.39
>    clat (msec): min=31, max=129, avg=46.94, stdev= 7.24
> 
> clat output of gnuplot reports units in msec. The clat gnuplot output 
> show a msec label on the y axis but trace some usec.

The fio_generate_plots doesn't take the usec vs msec into account, it's
basically hard wired. If someone wants to improve the script for better
reporting I'd be happy to take it, it's mainly considered an example of
how to do the reporting quickly.

> slat cli reports values that doesn't match at all what is viewable from 
> gnuplot output. Averagage is said to be 16 on the cli whereas the 
> gnuplot trace shows 20 at min.
> Max is said to be 145 for slat but displayed at 60 on the gnuplot.
> Slat is also said to be usec on the cli but shown as msec on the y axis 
> of the gnuplot.
> 
> Complete archive showing that could be downloaded at 
> http://konilope.linuxeries.org/tmp/localr.tar.bz2
> 
> It seems that units may vary from the cli output regarding the range of 
> results. I have some case where clat is shown in usec instead of msec. 
> Maybe that what confuse the gnuplot output.
> Like in :
>  write: io=32,003MiB, bw=124MiB/s, iops=1,294, runt=263681msec
>    slat (usec): min=12, max=71, avg=15.43, stdev= 2.72
>    clat (usec): min=663, max=1,537K, avg=49431.63, stdev=108787.82
>    bw (KiB/s) : min=    0, max=171311, per=99.91%, avg=127152.93, 
> stdev=21619.46

I'll take a look at this.

-- 
Jens Axboe

Re: Inconsency between cli output and gnuplot

[Jens Axboe]

On Tue, Dec 09 2008, Jens Axboe wrote:
> > slat cli reports values that doesn't match at all what is viewable from 
> > gnuplot output. Averagage is said to be 16 on the cli whereas the 
> > gnuplot trace shows 20 at min.
> > Max is said to be 145 for slat but displayed at 60 on the gnuplot.
> > Slat is also said to be usec on the cli but shown as msec on the y axis 
> > of the gnuplot.
> > 
> > Complete archive showing that could be downloaded at 
> > http://konilope.linuxeries.org/tmp/localr.tar.bz2
> > 
> > It seems that units may vary from the cli output regarding the range of 
> > results. I have some case where clat is shown in usec instead of msec. 
> > Maybe that what confuse the gnuplot output.
> > Like in :
> >  write: io=32,003MiB, bw=124MiB/s, iops=1,294, runt=263681msec
> >    slat (usec): min=12, max=71, avg=15.43, stdev= 2.72
> >    clat (usec): min=663, max=1,537K, avg=49431.63, stdev=108787.82
> >    bw (KiB/s) : min=    0, max=171311, per=99.91%, avg=127152.93, 
> > stdev=21619.46
> 
> I'll take a look at this.

It's weird, must be some gnuplot oddity. The data points are definitely
there and the range of the y-axis reflects this as well, however it
seems to be missing in the output. I've tried increasing the resolution,
but it doesn't change this fact.

-- 
Jens Axboe