Re: [gentoo-user] Good file system that recovers from a power failure.

[William Kenworthy]

Reply inline

On Sun, 2011-01-02 at 21:06 -0500, Walter Dnes wrote:
> On Sun, Jan 02, 2011 at 10:31:42AM +0800, William Kenworthy wrote
>
> > I use "dirvish" for backups which creates a LOT of hardlinks which can
> > be very hard on a file system.  ext2 typically lasts only a few cycles,
> > while ext3 is only a little better even with full journalling.  Coupled
> > to the fact neither is very good with power cuts and they are a worst
> > case choice for data security :)

>   Am I mis-understanding or are you mis-speaking?  hardlinks != backup
> A hardlink is simply another pointer to the same tracks/sectors on disk.
> If the on-disk data is destroyed it doesn't matter how many pointers you
> have to the data, it's gone.  A real backup is another copy of the data
> on another drive, preferably external.
>

Yes you have misunderstood, check out http://www.dirvish.org/.  Basicly
the first backup (--init) is a complete copy of the source into either a
local disk or remote storage.  Subsequent backups create a new image, by
checking if the previous copy of a file/directory/whatever has changed
and if not it will create a hardlink, but if changed will make a new
copy.  So you can have full, daily backups using typically only 2x the
original space for many versioned backups.  As only changed files are
copied, its only changes that use "real" space.

Restore is just copy the version back you want.  Full OS restore is done
in a similar fashion to copying one system to another (i.e., cloned from
the image).



> > Reiserfs3 by contrast is very very good, with only a few instances of
> > problems over many years (since beore 3 was even in the kernel) - none
> > of which have lost critical data or file systems (ext2/3 devs, are you
> > listening :)
>
>   I don't think ext2fs is being developed as such.  And ext3fs is mostly
> a journalling system backported to ext2fs.  ext2fs was written way back
> when in January 1993, and the specs were uptodate for then, but our
> expectations, and disk sizes have grown since then.
>
> > So, for me at least, btrfs is looking like the way forward.  Its in
> > "testing" at the moment, but I am ready to move whole systems over
> > to it.
>
>   I'm on reiserfs3 for now.  Hopefully, it'll be maintained until ext4
> or btrfs or whatever is deemed ready for primetime.  When that happens,
> I'll do any new installs on the new filesystem.  If it works, don't muck
> around with it.  Unless support/maintenance for reiserfs3 is dropped or
> a new fs comes out with a feature I really want/need, I won't migrate
> existing systems.
>
Exactly, I have had great service from reiserfs3, but fscking terrabytes
of storage is becoming a serious limitation when it means taking a
system offline to do so.  That being said, I only do it every 6 months
or so as a precaution rather than the expectation of finding something
wrong - and haven't unless it was an actual disk failure (that one was
at least 18 months ago.

BillK