RAID VFS module SoC project

Gerald (Jerry) Carter jerry at
Sat May 27 19:28:30 GMT 2006

Hash: SHA1

Hey Ming,

I was thinking about how an administrator would approach
this feature.  From a development standpoint I think
it would be good to start with implementing simple directory
tree mirroring initially to get a grip on how the VFS
layer works.  Of course, all of this is above the file
system layer I'm assuming.  If can think of how to do
it otherwise, I'd be interested.  But to be portable
(which I think is a real plus), it would have to be
file system agnostic.

The configuration might look something like (assuming
the library is named

    raid:level = 1
    raid:disks = /data/one /data/two

In this case, the directory tree in /data/two would
mirror that of /data/one.  If these were physically
seaparate disks then the admin has achieved some
level of fail over.  If the disk mounted at /data/one
dies, then the share can be manually switched to
provide access to /data/two.

The next question is how does this relate to the
path?  Is the top level directory for the share
just an empty folder on disk and the real data
stored in the 'raid:disks' path values?

Building the initial mirror is a little bit more work.
I'd accept using rsync for establishing the initial
mirror (outside of smbd).  If we can be more clever
later on, that good.  But at least we can get started.

We can look at striping later.  Level 1 probably makes
more  sense initially.  Raid 0+1 might be interesting.
I'm not how to really leverage the advantages of RAID 5.

cheers, jerry
Samba                                    -------
Centeris                         -----------
"What man is a man who does not make the world better?"      --Balian
Version: GnuPG v1.4.2 (GNU/Linux)
Comment: Using GnuPG with SUSE -


More information about the samba-technical mailing list