[clug] Managing NAS units?

Robert Edwards bob at cs.anu.edu.au
Fri Nov 2 02:52:16 GMT 2007


Hi Chris,

So, sounds like you need a single filesystem with multiple clients
accessing it. You can do this through a single file-server (back to
your original suggestion). This is what NFS and SMB are good at.

Alternatively, you have a clustered filesystem shared to multiple
file-servers (NFS, SMB, other, all of the above). The file-servers
see the NAS devices as "raw" block devices using iSCSI, AoE, Network
Block Device (NBD), Fibre-Channel etc.

Suitable clustered filesystems include GFS (have played with) or
Lustre (now owned by Sun, I believe) as well as IBMs GPFS or Suns
ZFS (SGI have one as well - can't remember the name).

For a big setup, I'd look at the second option. Yes, there is some
more indirection involved (increased overheads etc.) but the fault-
tolerance and scalability of the system should pay for itself in
the longer term.

Cheers,

Bob Edwards.

Chris Smart wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Robert Edwards wrote:
>> My question is: why did those quite independent services need to
>> share the NAS? Why couldn't they each have a smaller NAS for primary
>> storage and possibly share a separate storage system for secondary
>> (backup) storage?
>>
> 
> Thanks everyone for your replies.
> 
> The purpose of the space is to house a data repository, which multiple
> computers (some without fibre connections) will write to.
> 
> So is the suggestions that I should just let each computer access them
> as a device on the network directly?
> 
> Firstly, computers without fibre won't be able to transmit data to the
> NAS units without going through a share. This could be solved by adding
> a fibre card I guess.
> 
> But then we have issues like Windows clients will see these as a dozen
> different drive letters, with a file system they won't undertsand.
> 
> Linux clients will see multiple devices but at least understand the file
> system.
> 
> So when it comes to writing massive amounts of data to the SAN, they
> have to fill up one device, then switch to another mount point when that
> gets full? Sounds ridiculous to me (you'll have say 10 x 4TB devices).
> 
> I guess I just want a simple way to present all NAS units as a single
> device and filesystem to the rest of the network..
> 
> - -c
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.6 (GNU/Linux)
> Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
> 
> iD8DBQFHKov8hZNk0P/rW0sRAqsSAKCSDG0CautEaZ2ild/U0VU8YrYAjwCeIaDp
> w+13+sAeqfu49tCLkUAdmMA=
> =rIs6
> -----END PGP SIGNATURE-----



More information about the linux mailing list