Re: Fragmentation (was: XFS NFS server Oops)

[Steve Lord <lord@sgi.com>]

On Tue, 2002-03-19 at 19:37, Federico Sevilla III wrote:
> On 19 Mar 2002 at 16:47, Steve Lord wrote:
> > xfs_db -r /dev/xxx (it can be mounted)
> > xfs_db: frag -f
> 
> Cool. (Obviously I never really dug into the xfs_db manpage before to find
> stuff like this out.) That seemed harmless so I checked out all my
> filesystems.
> 
> I got fragmentation factors from a high of 50.92% (this is a filesystem
> where I store huge tar.gz backups and not much else) to a low of 0.43% for
> my / filesystem.
> 
> So now I wonder (trivial, really, but this might be FAQ-worthy):
> 
> a. At what fragmentation factor should we start considering running
> xfs_fsr? Or should we just go with what the xfs_fsr manpage says and run
> it in a crontab weekly?
> 
> b. I remember awhile back that xfs_fsr didn't come very highly
> recommended. Is this still the case? Or can we use xfs_fsr on production
> systems (during their more idle times, of course) and still be able to
> sleep at night?
> 

Probably hold off for now on running fsr.

Look at the other numbers on the output. The actual and ideal are
more interesting. If you look at these, the difference is the number
of extra extents you have above the ideal case. Then ask how much
data you have on the disk, dividing by the actual extent number
gives you the average length of the extents. It is also possible
that most of the fragmentation is restricted to a few files.

Steve

-- 

Steve Lord                                      voice: +1-651-683-3511
Principal Engineer, Filesystem Software         email: lord@sgi.com