[Gllug] Re: RAID on laptop or xfs?

[Liam Smit]

Well that does sound painful. I've had problems with this sort of
thing but only before changing to ReiserFS (from EXT2) but then I also
aquired a UPS for the machine in question...

Of course a laptop has a battery so don't need a UPS, which is good
because the Lead / Acid batteries in UPS weigh a _lot_ .<g>

My understanding is that the integrity of the journal itself is not
that crucial fro the integrity of the filesystem because as soon as
the journal gets hosed by the drive losing power and the contents of
the cache dissapearing into the ether, the journal is no longer in a
postition to write its contents to the rest of the file system.

It could be interesting to see the effect that different cache sizes
(512K, 2MB, 8MB) and journal sizes have on the problem. Of course if
the drive is busy making the changes that the journal has stroed int
it to the rest of the FS atm the power is lost then there could well
be corruption, if the power loss also hoses the journal at the same
time then undoing any changes woudl be far more difficult.  Then it' s
time to say fsck it and / or reach for the backups.

cheers
Liam


On Fri, 1 Apr 2005 07:04:13 +0100, Bruce Richardson
<itsbruce at uklinux.net> wrote:
> On Fri, Apr 01, 2005 at 01:33:44AM +0100, Liam wrote:
> > 
> > I'm not sure how much hdd corruption you will suffer if you are using
> > a journalling file system when the drive loses power, all changes to
> > files being stored in the journal.
> 
> Except that they will not all be there if some of the changes were in
> the writeback cache, that's the whole problem.  What makes it worse is
> that these drives do not usually flush the data blocks from the cache to
> the disk in the same order that they were written to it.  This can make
> complete nonsense of a journal or mean that only random portions of a
> RAID stripe were written to disk at poweroff.
> 
> > 
> > If some of that cache is for reads (i.e. not stuff waiting to be
> > written) than you can ignore that for data corruption purposes.
> 
> It's called a writeback cache.
> 
> -- 
> Bruce
> 
> Explota!: miles de lemmings no pueden estar equivocados.
> 
>
-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug