I have a pile of aging Dell MD-1000''s laying around that have been replaced by new primary storage. I''ve been thinking of using them to create some archive/backup storage for my primary ZFS systems. Unfortunately they do not all contain identical drives. Some of the older MD-1000''s have 15x500GB drives, some have all 750''s some all 1TB''s. Since size and integrity matters here, not speed. I was thinking of creating one large pool containing multiple RAIDZ2''s. Each RAIDZ2 would be one MD-1000 and would have 14 drives, reserving one drive per shelf as a spare. The question is: The final pool would have spares of 500GB, 750GB and 1TB. Is ZFS smart enough to pick the right one if a drive fails? If not, is there a way to make this scenario work and still combine all available storage in a single pool? Willem -- This message posted from opensolaris.org
On Jan 12, 2011, at 5:45 PM, Wim van den Berge wrote:> I have a pile of aging Dell MD-1000''s laying around that have been replaced by new primary storage. I''ve been thinking of using them to create some archive/backup storage for my primary ZFS systems. > > Unfortunately they do not all contain identical drives. Some of the older MD-1000''s have 15x500GB drives, some have all 750''s some all 1TB''s. Since size and integrity matters here, not speed. I was thinking of creating one large pool containing multiple RAIDZ2''s. Each RAIDZ2 would be one MD-1000 and would have 14 drives, reserving one drive per shelf as a spare.If you are going to put all drives in a shelf into a single vdev, then it will be better to use raidz3 than raidz2+spare.> > The question is: The final pool would have spares of 500GB, 750GB and 1TB. Is ZFS smart enough to pick the right one if a drive fails? If not, is there a way to make this scenario work and still combine all available storage in a single pool?Use warm spares instead of hot spares, or raidz3. -- richard
On Wed, Jan 12, 2011 at 5:45 PM, Wim van den Berge <wvandenberge at altep.com> wrote:> I have a pile of aging Dell MD-1000''s laying around that have been replaced by new primary storage. I''ve been thinking of using them to create some archive/backup storage for my primary ZFS systems. > > Unfortunately they do not all contain identical drives. Some of the older MD-1000''s have 15x500GB drives, some have all 750''s some all 1TB''s. Since size and integrity matters here, not speed. I was thinking of creating one large pool containing multiple RAIDZ2''s. Each RAIDZ2 would be one MD-1000 and would have 14 drives, reserving one drive per shelf as a spare. > > The question is: The final pool would have spares of 500GB, 750GB and 1TB. Is ZFS smart enough to pick the right one if a drive fails? If not, is there a way to make this scenario work and still combine all available storage in a single pool?While it may not be recommended as a best practise, there''s nothing "wrong" with using vdevs of different sizes. You can even use vdevs of different types (mirror + raidz1 + raidz2 + raidz3) in the same pool, although you do have to force (-f) the add command. My home ZFS box uses a 3-drive raidz1 vdev and a 2-drive mirror vdev in the same pool, using 160 GB SATA and 120 GB IDE drives. My work storage boxes use 8-drive raidz2 vdevs, mixed between 0.5 TB SATA, 1.0 TB SATA, and 1.5 TB SATA. Performance won''t be as good as it could be due to the uneven striping, especially when the smaller vdevs get to be full. But it works. -- Freddie Cash fjwcash at gmail.com