redundant storage

Fri Jun 3 12:10:47 UTC 2016

On Fri, Jun 03, 2016 at 11:47:46AM +0100, Steve O'Hara-Smith wrote:
> On Fri, 3 Jun 2016 12:14:46 +0200
> Julien Cigar <julien.cigar at gmail.com> wrote:
> 
> > On Fri, Jun 03, 2016 at 10:41:38AM +0100, Steve O'Hara-Smith wrote:
> > > 	Hi,
> > > 
> > > 	Just one change - don't use RAID1 use ZFS mirrors. ZFS does
> > > better RAID than any hardware controller.
> > 
> > right.. I must admit that I haven't looked at ZFS yet (I'm still using
> > UFS + gmirror), but it will be the opportunity to do so..!
> > 
> > Does ZFS play well with HAST?
> 
> 	Never tried it but it should work well enough, ZFS sits on top of
> geom providers so it should be possible to use the pool on the primary.
> 
> 	One concern would be that since all reads come from local storage
> the secondary machine never gets scrubbed and silent corruption never gets
> detected on the secondary. A periodic (say weekly) switch over and scrub
> takes care of this concern. Silent corruption is rare, but the bigger the
> pool and the longer it's used the more likely it is to happen eventually,
> detection and repair of this is one of ZFSs advantages over hardware RAID
> so it's good not to defeat it.

Thanks, I'll read a bit on ZFS this week-end ..!

My ultimate goal would be that the HAST storage survives an hard reboot/
unplugged network cable/... during an heavy I/O write, and that the
switch between the two nodes is transparent to the clients, without any
data loss of course ... feasible or utopian? Needless to say that what 
I want to avoid at all cost is that the storage becomes corrupted and
unrecoverable..!

> 
> 	Drive failures on the primary will wind up causing both the primary
> and the secondary to be rewritten when the drive is replaced - this could
> probably be avoided by switching primaries and letting HAST deal with the
> replacement.
> 
> 	Another very minor issue would be that any corrective rewrites (for
> detected corruption) will happen on both copies but that's harmless and
> there really should be *very* few of these.
> 
> 	One final concern, but it's HAST purely and not really ZFS. Writing
> a large file flat out will likely saturate your LAN with half the capacity
> going to copying the data for HAST. A private backend link between the two
> boxes would be a good idea (or 10 gigabit ethernet).

yep, that's what I had in mind..! one nic for the replication between
the two HAST node, and one (CARP) nic by which clients access to
storage..

> 
> > > On Fri, 3 Jun 2016 10:38:43 +0200
> > > Julien Cigar <julien.cigar at gmail.com> wrote:
> > > 
> > > > Hello,
> > > > 
> > > > I'm looking for a low-cost redundant HA storage solution for our
> > > > (small) team here (~30 people). It will be used to store files
> > > > generated by some webapps, to provide a redundant dovecot (imap)
> > > > server, etc.
> > > > 
> > > > For the hardware I have to go with HP (no choice), so I planned to buy
> > > > 2 x HP ProLiant DL320e Gen8 v2 E3-1241v3 (768645-421) with 
> > > > 4 x WD Hard Drive Re SATA 4TB 3.5in 6gb/s 7200rpm 64MB Buffer 
> > > > (WD4000FYYZ) in a RAID1 config (the machine has a smartarray P222
> > > > controller, which is apparently supported by the ciss driver)
> > > > 
> > > > On the FreeBSD side I plan to use HAST with CARP, and the volumes
> > > > will be exported through NFS4.
> > > > 
> > > > Any comments on this setup (or other recommendations) ? :)
> > > > 
> > > > Thanks!
> > > > Julien
> > > > 
> > > 
> > > 
> > > -- 
> > > Steve O'Hara-Smith <steve at sohara.org>
> > > 
> > 
> 
> 
> -- 
> Steve O'Hara-Smith                          |   Directable Mirror Arrays
> C:>WIN                                      | A better way to focus the sun
> The computer obeys and wins.                |    licences available see
> You lose and Bill collects.                 |    http://www.sohara.org/
> 

-- 
Julien Cigar
Belgian Biodiversity Platform (http://www.biodiversity.be)
PGP fingerprint: EEF9 F697 4B68 D275 7B11  6A25 B2BB 3710 A204 23C0
No trees were killed in the creation of this message.
However, many electrons were terribly inconvenienced.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <http://lists.freebsd.org/pipermail/freebsd-questions/attachments/20160603/bac269cd/attachment.sig>