xfs partition layout and using xfs_growfs

xfs partition layout and using xfs_growfs

[Linda A. Walsh]

Is there any difference in a file systems layout and efficiency between
to xfs fs's of the same size, but where one was created at size '100%', 
where the other was created at size 50%, but then grown iteratively 
to 60, 70, 80, 90 and then 100% over time as it filled?

Would the final file systems look pretty much the same and have roughly the
same performance characteristics?  Assume, for sake of argument, that the
file system was grown before space got tight enough to cause any severe 
large file fragmentation.  

I've been under the impression that one gained some performance benefits if one
laid out the whole file system at once  is that a mis-impression?

That asked/said...is there any work underway to create an xfs_shrinkfs, so
that one could go the other way?

Thanks,
Linda

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: xfs partition layout and using xfs_growfs

[Dave Chinner]

On Sun, Jun 20, 2010 at 09:35:13PM -0700, Linda A. Walsh wrote:
> 
> Is there any difference in a file systems layout and efficiency between
> to xfs fs's of the same size, but where one was created at size
> '100%', where the other was created at size 50%, but then grown
> iteratively to 60, 70, 80, 90 and then 100% over time as it filled?

Yes.

> Would the final file systems look pretty much the same

>From a structural perspective, it depends on the size of the AGs in
the initial filesystem.  If the filesystem was big enough to begin
with (i.e.  using maximally sized AGs at mkfs time) then growing it
will result in a layout exactly the same as if it was mkfs'd at full
size.

However, the allocation patterns will be significantly different
between the two filesystems, so there will be no similarity in data
layout between the two different filesystems.

> and have roughly the
> same performance characteristics?

See previous answer.

> Assume, for sake of argument, that the
> file system was grown before space got tight enough to cause any
> severe large file fragmentation.

See previous answer. ;)

> I've been under the impression that one gained some performance benefits if one
> laid out the whole file system at once  is that a mis-impression?

I wouldn't say you "gain performance benefits" by starting with a
larger filesystem, more that a grown filesystem has different aging
characteristics to one that has not been grown. See previous answer.
;)

> That asked/said...is there any work underway to create an xfs_shrinkfs, so
> that one could go the other way?

http://xfs.org/index.php/Shrinking_Support

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs