definitions for /proc/fs/xfs/stat

[Mark Seger]

I'm doing 1 second samples and the rates are very steady.  The reason I
ended up at this level of testing was I had done a sustained test for 2
minutes at about 5MB/sec and was seeing over 500MB/sec going to the disk,
again sampling at 1-second intervals.  I'd be happy to provide detailed
output and can even sample more frequently if you like.

from my shorter test I was experimenting looking at some of the XFS data
with collectl and recorded this, it if help at all:

segerm at az1-sw-object-0006:~$ collectl --import xfs
waiting for 1 second sample...
#<--XFS Ops--><-----------XFS
Logging----------><------Extents------><------DirOps-------><----Trans---><-
# Write Reads  Writes WrtKBs NoRoom Force Sleep  ExtA BlkA ExtF ExtF  Look
Cre8 Remv Gdnt  Sync Asyn Empt
      0     0       0      0      0     0     0     0    0    0    0     0
   0    0    0     0    0    0
      0     0       3    768      0     3     3     0    0    0    0     0
   0    0    0     3    0    0
      0     0       0      0      0     0     0     0    0    0    0     0
   0    0    0     0    0    0
      0     0       0      0      0     0     0     0    0    0    0     0
   0    0    0     0    0    0
      0     0       0      0      0     0     0     0    0    0    0     0
   0    0    0     0    0    0
      0     0       1    256      0     1     1     0    0    0    0     0
   0    0    0     1    0    0
     38    88      95  24320      0    95    96    54   54   54   54   115
  76   76  154    95  473    0
    339   776     968 247816      0   968   978   484  484  479  479  1011
 675  671 1351   967 4087    0
    321   748     929 237806      0   929   935   450  450  453  453   967
 645  647 1287   930 3847    0
    279   637     810 207360      0   810   811   391  391  390  390   838
 559  558 1118   810 3324    0
    209   482     610 156160      0   610   610   286  286  289  289   627
 417  420  834   610 2451    0
      0     0       3    768      0     3     3     0    0    0    0     0
   0    0    0     3    0    0

I can say for a fact I was doing about 300 wtrites/sec whcih the write
numbers seem to support, though I don't know what the read numbers are
measuring.  you can also see from the logging data that was 250MB/sec going
to disk.

Are there other numbers that are meaningful that you'd like to see?  All it
takes is adding a couple of print statement as what you're seeing above
only took a hour or so to throw to together.

I can collect as much (or little as you like) and actually can save the
complete contents of /proc/fs/xfs/stat every second in a file for later
playback.

-mark



On Fri, Jun 14, 2013 at 8:17 PM, Nathan Scott <nathans at redhat.com> wrote:

>
>
> ----- Original Message -----
> > actually I have since found a decoder ring here -
> > http://xfs.org/index.php/Runtime_Stats and have been incorporating a
> lot of
> > the data so I can look at things in real time.  I'd still love to know
> why
>
> Ah, good stuff.
>
> > writing 1000 1K files results in 200MB/sec of disk I/O though.  clearly
>
> For how many seconds?  (or fractions thereof?)  You may need the level of
> detail that only tracing will provide to explore further, although off the
> top of my head I'm not sure exactly which events you should focus on (log,
> inode creation, or space allocation at a guess).
>
> cheers.
>
> --
> Nathan
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://oss.sgi.com/pipermail/xfs/attachments/20130614/309f8e66/attachment.html>