On Tuesday 11 May 2010 15:40:09 Sorin Srbu wrote:
> I think the culprit was the software raid5-array the whole time... Peaking
>  at approx 94-95% full pool since a few months back, the array was
>  stretched nearly to the max for all intents and purposes.

RAID-5, in my experience, is not fast in software or hardware.  Write 
operations are still O(1), only read is improved. If you can get 4 disks, do 
RAID-10.

However, there is another consideration. When your filesystem was 95% full, did 
you have some part of the filesystem reserved? When normally creating a 
filesystem, this is set at 5%, which means "95%" means "95% of the way to 
hitting the reserved space".

If you had no reserved space, and the filesystem was _actually_ at 95% full, 
that could also be a source of the problem.  I've not experienced it 
personally, but I've heard it said that ext filesystems do not deal well with 
being very close to full. Once there isn't much space left, new files become 
increasingly fragmentated, as it isn't possible to find consecutive free blocks 
on the disk.

I'm in the process of eliminating all the RAID-5 in my organization. RAID-1 or 
RAID-10 both perform far better, and they don't have the "write hole" problem 
(which I have personally been victim of, when a power supply failed 
spectacularly).

Regards,
Tyler

-- 
"Standard creepiness rule: don't date under (AGE/2 + 7)."
   -- Randall Munroe, XKCD, http://xkcd.com/314/

------------------------------------------------------------------------------

_______________________________________________
BackupPC-users mailing list
BackupPC-users AT lists.sourceforge DOT net
List:    https://lists.sourceforge.net/lists/listinfo/backuppc-users
Wiki:    http://backuppc.wiki.sourceforge.net
Project: http://backuppc.sourceforge.net/