01-01-2019, 09:32 PM
This can easily be determined: RAID is not a backup. If the RAID system is so bad that it hits the wall, then hopefully you have a backup. The maximum amount you can lose depends on how often you back up your crate.
Back to the topic: RAID protects against one thing in particular: defects in RAID components. Especially against defective mass storage devices. If you look at the old comparison to the car, then the hard disks are something like brake pads or clutch discs. So wear parts. They break from time to time and have to be replaced.
The aim of the RAID is to keep the box running while it waits for the hardware to be replaced. For the defect of the RAID (i.e. the condition that the RAID itself is dismounted and not just a component) you have a backup.
So the question is which RAID will give you what you need in case of a LW defect. And the RAID-1 definitely has an advantage because it is a 1:1 mirror. So in case of an error you only have a reduction of the bandwidth for read accesses (because only half of the disks can be read - e.g. two instead of four).
The RAID-5 must replace the missing disk by calculating the missing data blocks from the parity blocks. This is associated with considerably lower performance in the event of a drive failure. I had once maintained a root server for a customer that was still running when a RAID-5 disk failed, but the performance was so poor that it was actually offline until both the disk had been replaced and the RAID had finished the resync. With a RAID-1, the server would probably have continued to function fully.
RAID-5 is then also much more resource-intensive when restoring the RAID with a new disk, since blocks do not simply have to be copied, but the parity blocks are recalculated.
RAID-1 and RAID-5 have a few other differences, but they have less to do with the error than with how many disks you write to at the same time (because Stripe-Set writes to distributed disks and Mirror has no advantage over a single disk in writing), but that has little to do with the failure mode behavior.
Back to the topic: RAID protects against one thing in particular: defects in RAID components. Especially against defective mass storage devices. If you look at the old comparison to the car, then the hard disks are something like brake pads or clutch discs. So wear parts. They break from time to time and have to be replaced.
The aim of the RAID is to keep the box running while it waits for the hardware to be replaced. For the defect of the RAID (i.e. the condition that the RAID itself is dismounted and not just a component) you have a backup.
So the question is which RAID will give you what you need in case of a LW defect. And the RAID-1 definitely has an advantage because it is a 1:1 mirror. So in case of an error you only have a reduction of the bandwidth for read accesses (because only half of the disks can be read - e.g. two instead of four).
The RAID-5 must replace the missing disk by calculating the missing data blocks from the parity blocks. This is associated with considerably lower performance in the event of a drive failure. I had once maintained a root server for a customer that was still running when a RAID-5 disk failed, but the performance was so poor that it was actually offline until both the disk had been replaced and the RAID had finished the resync. With a RAID-1, the server would probably have continued to function fully.
RAID-5 is then also much more resource-intensive when restoring the RAID with a new disk, since blocks do not simply have to be copied, but the parity blocks are recalculated.
RAID-1 and RAID-5 have a few other differences, but they have less to do with the error than with how many disks you write to at the same time (because Stripe-Set writes to distributed disks and Mirror has no advantage over a single disk in writing), but that has little to do with the failure mode behavior.