> From: zfs-discuss-boun...@opensolaris.org [mailto:zfs-discuss- > boun...@opensolaris.org] On Behalf Of John Andrunas > > I know this is potentially a loaded question, but what is generally > considered the optimal disk configuration for ZFS. I have 48 disks on > 2 RAID controllers (2x24). The RAID controller can do RAID > 0/1/5/6/10/50/60 or JBOD. What is generally considered the optimal > configuration for the disks. JBOD, RAID zvols on both controllers.
No matter what your goals are, you should JBOD the disks, and let ZFS manage the raid. It not only performs better than hardware raid, it's also more reliable. When you scrub, ZFS will be able to access all the bytes on all the disks. But if you had something like a hardware mirror, the hardware would only present a single device to the OS, and therefore the OS wouldn't be able to scrub all the bits on both sides of the mirror. If there's a data error encountered in hardware mirror, ZFS has no 2nd device to read from to correct the error. All it can do is blindly retry the hardware, and hope for a different answer. Which probably isn't going to happen. Generally speaking, your choices are: Stripe, Mirror, Stripe & Mirror, or Raid(z,z2,z3) Generally speaking, a simple stripe gets you the highest performance and capacity per dollar. But no redundancy, so your whole pool is at risk. Very few scenarios can accept this sacrifice of reliability to gain the incremental performance. Generally speaking, a n-way mirror writes at the speed of 1 disk, reads at the speed of n-disks, and costs (n)x a single disk, for the capacity of a single disk. Generally speaking, a stripe of mirrors (suppose 3-way stripe of 2-way mirrors, totaling 6 disks) writes 3 times faster than a single disk, reads 6x faster than a single disk, costs 6 disks, and has capacity of 3 disks. Generally speaking, raid has the capacity of n-1, n-2, or n-3 disks depending on which config you choose. It performs just as fast (n-1, n-2, or n-3) for large sequential operations. But the performance for small random operations is poor. Maybe 1 disk or 2 disks performance. Raid is usually what you use when you need high capacity, and you need redundancy, and you're budget constrained. You are forced to sacrifice some speed. You could have gotten better performance by striping mirrors, but then you'd have to spend more money on disks to have the same usable space. _______________________________________________ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
