"Stateless Linux" project

Per Bjornsson perbj at stanford.edu
Tue Sep 14 23:31:13 UTC 2004


On Tue, 2004-09-14 at 14:02, Havoc Pennington wrote:
> On Tue, 2004-09-14 at 13:40 -0400, Bryan K. Wright wrote:
> > 	On another topic, the overview notes that disaster recovery for
> > laptops requires that the laptop home directory be backed up often --
> > "perhaps whenever the laptop connects to the intranet".  Toward this end,
> > will the stateless project incorporate support for network block devices?
> > A local disk in a "RAID 1" array with a remote block device sounds like
> > a good way to ensure that an up-to-date remote copy of files is maintained
> > while the user is connected.  When the connection is broken, the remote
> > half of the mirror would be unavailable, and the local disk would be used.
> > On reconnection, the halves of the mirror would resync.
> 
> That's a clever idea. Dan Reed has some code written that's a bit
> simpler approach - it just rsyncs the homedir periodically.

Sounds like it would actually functionally be pretty similar? At least
if the rsync is done "as soon as possible" when the notebook gets
connected to the "home" network.

Actually I think this really needs some more smarts, in order to be
really useful it has to be able to deal with changes made both on the
server and on the client (at least if shared directories are included,
and even if they aren't if one considers the case of one user who uses
several computers using the same home directory). Possibly Unison or
something similar might be a starting point:
http://www.cis.upenn.edu/~bcpierce/unison/

Such a system keeps track of what files got changed on what side. It
pretty much punts to the user if changes have been made to a file on
both ends. (You can plug in a merge tool for merging text files though;
I don't think that there is really any hope to "merge" more complex
documents with any particular degree of automation.)

I keep directories that I'm working in synced between a file server and
my notebook, and I do sometimes muck around in both sides of the
connection (e.g. dumping new measurement data on the file server,
syncing, and analyzing the data on my notebook). This kind of serves as
a simple userspace cached network filesystem.

> The hard questions on homedir backup seem to be around how it interfaces
> to an "industrial strength" backup solution, i.e. we can keep the
> homedir synced to a network share pretty easily, but how does that
> interact with incremental backups and so forth.

Well backing up the disconnected notebook on a main backup server sounds
hard! ;) In what scenario does just backing up the server side fail
(apart from losing any changes since the last check-in)? As far as I can
tell the backups should never really change any file state (apart from
possibly atime) on the files?

/Per

-- 
Per Bjornsson <perbj at stanford.edu>
Ph.D. Candidate, Department of Applied Physics, Stanford University





More information about the fedora-devel-list mailing list