[Orca-users] VXFS fragmentation

[Liston Bias]

On Fri, 24 Jan 2003, Lightfoot.Michael wrote:

> > We have identified a bug in VXFS 3.4 that is not allowing us
> > to reaname a file if it happens to exist in directory with
> > fragmented directory inode. We are working that issue with
> > Veritas... but it got us thinking.
> >
> I'd like to read up on this, where is it docuemnted on the Sun or
> Veritas sites?  We haven't upgraded to 3.4 (it was planned but pushed
> into the background when other projects of a more urgent nature arose.)

They supposedly have a patch coming out the beginning of February that
will resolve the problem.  The also send us an interim patch 112375-01
that may solve some issues for us:

Topic: Point Patch for VxFS 3.4p3 large directory cache issue( i96374 )

I don't have a bugid for the incident but will check with Veritas to see
if there is something they provide me in the realm.

> > We don't ever run vxfs fsadm on our systems unless problem identified.
> >
> > In reasearching docs, there is recommendations ranging from
> > daily to monthly cron of this function on systems.  I know it
> > is I/O intensive to run these functions, so am looking to
> > "display" the fragmentation that actually exist to justify runs.
> >
> > Has anyone taking a shot at graphing fragmentation with orca?
> >
> It's probably a pointless exercise unless fragmentation changes rapidly.
> I run a frag report on our two systems with VxFS every Sunday morning at
> about 3am.  I have run a defrag once in 12 months and only because I
> wanted to check that the defrag script worked (so I could complete the
> documentation for the systems.)
>
> One of the advantages of VxFS is that it doesn't fragment very quickly,
> even on busy systems, unlike UFS which can go bad very quickly.  I have
> two squid proxy caches which are currently on UFS and they are already
> (after less than 3 months) getting significantly fragmented (I don't
> graph this with Orca either - just run a daily emailed frag report using
> fsck -n.)
>
> At another site I workjed at nearly two years ago I convinced management
> to fork out the dosh to put all my top tier squid proxy caches on VxFS.
> As far as I know they only need to defrag these every six months or so
> (they all have 8GB cache directories, some with two and some with four)
> unlike the previous UFS ones that caused problems (one actually ran out
> of free blocks) after less than 4 months.

The server in question has about 1.5 Terabytes of information on it and is
very active with creating, removing, renaming files for our billing
system.  Part of this activity is oracle but that doesn't seem to be the
problem volumes.

- Liston


============================================================================
- Liston Bias                       The really happy man is one who can
  Alumnus of Oklahoma State Univ    enjoy the scenery on a detour.
  Alumnus of Florida State Univ                     -- Anonymous

  bias at pobox.com
  http://www.pobox.com/~bias
============================================================================