[ale] Again with the filesystem recovery SOLUTION?

[Chris Ricker]

On Tue, 28 Jan 2003, Jason Day wrote:

> On Mon, Jan 27, 2003 at 11:03:03PM -0600, ChangingLINKS.com wrote:
> > Please give 3 benefits of ext3 over ext2 or xfs. I admit, that I usually take 
> > the quicker and easier way to the solution. I was only using ext3 so that I 
> > would not have to run fsck after rebooting after a crash - but now I know 
> > that the cost of ext3 is not worth the reward.
> 
> I'm no expert, so I can't give nitty-gritty details of how one
> filesystem might be better than another.  I have been using ext3 for
> over a year without problems, so I can tell you what I've observed.
> 
> One benefit of ext3 over ext2 is the one you mention above, no fsck
> after a crash.  I have also been under the impression that ext3 was
> safer than ext2; that is, if there is a crash, you are less likely to
> lose data with ext3 than with ext2.  I don't *know* if that's true, it's
> just the impression I have.  It is also my impression that if there is a
> problem with the filesystem, you are more likely to recover your data
> with ext3 than ext2.  Again, I don't know this for a fact.  So there's
> one, possibly three benefits of ext3 over ext2.

ext3 runs in three different modes:

data=journaled
data=write-back
data=ordered

data=ordered is the default. It gives you metadata journalling (so that you
don't need an fsck after a crash), and it also gives you a reordering of
transactions to ensure maximal integrity (conceptually similar to the soft
updates stuff McKusick added to FreeBSD UFS). There, ext3 is giving you more 
data protection than ext2, plus no fsck

data=journaled is a full data and metadata journal. You don't need an fsck, 
and you're getting insane data protection, particularly if you then do 
something like put the journal on a solid-state ram drive.

data=write-back kills the journalling entirely, so you still need an fsck,
but still reorders transactions for some data protection

later,
chris
_______________________________________________
Ale mailing list
Ale at ale.org
http://www.ale.org/mailman/listinfo/ale