On Sat, Jun 19, 2004 at 11:03:09PM -0600, J. Ryan Earl wrote:
> OK, as promised, have some more questions.
>
> I'm looking to build a cluster of diskless machines with failover and
> redundancy. I was wanting to run RAID1 on a network based filesystem.
> I was thinking I could possibly run RAID1 over NBD. Has anyone done
> this successfully?
Pavel (Machek, the original NBD author) has, but I don't know any
details.
> I'm looking for a situation where I have N servers with a harddrive in
> them, N > 1, let's call them masters. Servers without harddrives I'll
> call nodes. Let's say I have 3 masters and 9 nodes. I'm thinking each
> of the master nodes will require their own seperate root filesystem,
> with the 9 nodes booting off PXE (Pre-eXecution Environment?) via a
> dhcpcd option that mounts the root as an md device across three NBDs.
>
> Does this sound reasonable?
I'm not sure what you mean by "reasonable", but it sure does sound
possible.
> Are there any documents on this out there?
Not that I'm aware of.
You'll be interested in <http://borderworlds.dk/projects/#nr> for
setting up NBD on diskless clients, I think.
> I've searched google and not found a whole lot. One of the things I
> read, and I don't remember where, was that you could swap across NBD.
> Is this true?
Yes, but you need a patched kernel. AFAIK, those patches are no longer
maintained since I'm not well versed in kernel-space programming, and
the current kernel-space NBD maintainer doesn't work with those
patches. The patch is required to avoid deadlocks (which would occur
when a client tries to allocate memory to be able to make a call to the
server so that memory can be swapped out...)
--
EARTH
smog | bricks
AIR -- mud -- FIRE
soda water | tequila
WATER
-- with thanks to fortune
Attachment:
signature.asc
Description: Digital signature