I would very much appreciate some advice on this;
For our file- and mail servers we have been using mirrored raid-5
chassises, with disksuite and ufs. This has served us well, and the
el-cheapo raid-5 chassises have failed several times without any
downtime for our services.
We are now looking for a modern but simple-to-handle-and-not-to-
expensive
replacement for these servers. We still like the idea of having
raid-51 or similar security. By some reason that I haven''t gotten
yet, zfs doesn''t allow you to put "raids" upon each other,
like
mirrors/stripes/parity raids on mirrors/stripes/parity raids, in a
single pool. Maybe it is just not necessary since you can make pools
out of pools.
Now we could choose from:
1) mirroring all disks with disk suite and build a raidz (or possibly
raidz2) on those
2) creating two raidz[2] pools, creating a volume each on those, and
creating a third pool which is a zfs mirror of the volumes.
Alternative 2 is obviously nicer since there is only one mechanism
involved and that needs to be supervised and handled instead of two, and
the track record for disk suite''s ability to track disks that move and
other handling stuff could have been better. It would probably also
give us even a little more security, since hot spares and data disks
aren''t paired.
But which setup would give us better performance?
Are there other issues we should consider when choosing?
Thank you in advance for advice and hints!
Ragnar Sundblad
------------------------------------------------
Systems Specialist
Department of Computer Science and Communication
Royal Institute of Technology, Stockholm, Sweden
On Wed, Feb 18, 2009 at 2:47 PM, Ragnar Sundblad <ragge at csc.kth.se> wrote:> > I would very much appreciate some advice on this; > > For our file- and mail servers we have been using mirrored raid-5 > chassises, with disksuite and ufs. This has served us well, and the > el-cheapo raid-5 chassises have failed several times without any > downtime for our services. > > We are now looking for a modern but simple-to-handle-and-not-to-expensive > replacement for these servers. We still like the idea of having > raid-51 or similar security. By some reason that I haven''t gotten > yet, zfs doesn''t allow you to put "raids" upon each other, like > mirrors/stripes/parity raids on mirrors/stripes/parity raids, in a > single pool. Maybe it is just not necessary since you can make pools > out of pools. > > Now we could choose from: > 1) mirroring all disks with disk suite and build a raidz (or possibly > raidz2) on those > 2) creating two raidz[2] pools, creating a volume each on those, and > creating a third pool which is a zfs mirror of the volumes.Why not just use zfs to mirror your hardware raid-5 chassis? You get the hardware to manage disk failures, but have data redundancy provided by zfs which is where you want it. If you want random I/O performance, raidz isn''t a good choice. For most things, hardware raid ought to give you more IOPS. You mentioned mail and file serving, which isn''t an obvious match for raidz (which works better for capacity and throughput). -- -Peter Tribble http://www.petertribble.co.uk/ - http://ptribble.blogspot.com/
2009/2/18 Ragnar Sundblad <ragge at csc.kth.se>:> For our file- and mail servers we have been using mirrored raid-5 > chassises, with disksuite and ufs. This has served us well, and the> By some reason that I haven''t gotten > yet, zfs doesn''t allow you to put "raids" upon each other, like > mirrors/stripes/parity raids on mirrors/stripes/parity raids, in a > single pool.Is there any reason why you don''t want to use striped mirrors (i.e. stripes of mirrored vdevs, a.k.a raid 10) with online spares? This should provide high level of availability while greatly reducing downtime needed for resilvering in the event of disk failure. And if you''re REALLY paranoid you could go with 3-way or more mirror for each vdevs. Regards, Fajar