On 2011-12-30 16:55, V G wrote: > On Fri, Dec 30, 2011 at 1:28 PM, Joe Koberg wrote: >> May I suggest a ZFS based filesystem? >> >> FreeBSD, Solaris, OpenSolaris, and Linux via beta 3rd-party driver, >> support this advanced filesystem. I would trust Solaris or FreeBSD the m= ost. >> >> Every ZFS metadata block contains a checksum of the blocks it >> references. Every time ZFS reads a block, it verifies the checksum. If >> it's not right, it reads the copy from the {mirror/raid} and rewrites >> the original block (thus self-repairing). This is far more advanced than >> simply checking if the IO channel returned a read error, and then trying >> it on the other IO channel (RAID-1). Even if the IO channel returns no >> errors, the data could still be corrupted. In fact, a standard hardware >> RAID will be happy to copy corruption from one disk to another in the >> name of "recovery", as long as the disks don't report read errors. ZFS >> can identify incorrect data coming off the disk, something a hardware >> RAID controller or a software emulation of a RAID controller cannot. >> >> ZFS supports volumes of many mirrored sets (my preference), or groups of >> RAID with single, double, or triple parity. (Meaning one, two, or three >> drives can fail). >> >> ZFS also supports real scrubbing (reading every allocated block and >> comparing it to its checksum)... Which done regularly, will identify and >> self-repair disk errors in a very reliable way. It also supports >> in-place growth, zero-cost snapshots, replication, etc, etc.... >> >> On Solaris it supports transparent kernel-level Windows file service >> via CIFS/SMB, with ACLs et al. >> >> Joe Koberg, AE5NE >> joe@osoft.us >> > That sounds good. So you're saying go with a ZFS system on software > RAID rather than hardware RAID? > > How do you feel about btrfs? > Correct. Hardware RAID is "blind" to whats actually in the sectors. You=20 can do better from a reliability standpoint. ZFS un-layers the storage=20 stack, so the filesystem, volume manager, and storage devices all work=20 better together. For example: Outside of idling the disks and waiting for the controller=20 to "patrol read" the entire disk surface, how do you read-verify both=20 halves of a RAID-1 mirror? You can't just run through the device with=20 "dd" or whatever, because it will only read half the mirror. However,=20 the ZFS "scrub" command is intended for precisely this purpose. btrfs is promising but doesn't seem ready to deploy in production. Joe --=20 http://www.piclist.com PIC/SX FAQ & list archive View/change your membership options at http://mailman.mit.edu/mailman/listinfo/piclist .