Re: SCSI Disk/Controller advice please
On Mon, 1 Nov 2004 15:22:17 +0100, Frank Gevaerts <frank@gevaerts.be> wrote:
> On Mon, Nov 01, 2004 at 09:28:32AM +0800, Paolo Alexis Falcone wrote:
> > RAID 5 alleviates this by using parity information stored across the
> > disks - now it takes more than 1 disk failure for RAID 5 to fail.
>
> How does this change anything ? If you have one failed disk, and one
> disk containing unknown errors (the same case as your RAID1 example
> above), replacing the failed disk will lead to errors on the new disk.
The use of parity information in separate blocks for reads and writes
would just reduce the risk of that happening (prolonging the
inevitable?) as data information and parity information are
distributed across all disks in the array (RAID 1 won't contain parity
information, and is just a copy - data and all) . The disadvantage, of
course, with this setup, the controller design would be a lot more
complicated and subsequently would make the array reconstruction more
difficult unlike RAID 1 wherein it's guaranteed that you'd get a copy
of the other half of the mirror. Whether that remaining half of the
mirror already got checked for other errors that might've seeped in is
another matter absent of RAID 1 though as RAID 1 doesn't have a
provision for other information other than merely write data to the
other disk as well.
On an environment that's heavy on writes, RAID 5's overhead doesn't
really justify the costs. You'd be better off with RAID 1 for that.
--
Paolo Alexis Falcone
pfalcone@gmail.com
Reply to: