Must a Raid5 setup have identical drives and identical drive capacities?

A single RAID 5 set does require identical capacities but doesn't care about the drives. However it is possible (and not even unusual) to have multiple RAID 5 sets on an array. We do that here with our Clariion array.

By single "set" I mean something that would appear as a single disk so far as the OS is concerned but be comprised of multiple disks where the array is concerned. So as with the first array I ever worked with we had 20 drives and split that up into 4 different RAID 5 sets each comprised of 5 drives. The OS in that case saw these as if they were 4 very large luns (logical units seen as if they were 4 single disks).

If you're talking software RAID you can probably do it by partitioning the drives into equal size partitions then doing your RAID set using these equal size partitions. You'd have to be careful though. NOTE: You wouldn't want to make a RAID 5 set that had two different partitions from the same drive as it would then be possible that the data and its parity could end up on the same physical disk so that if you lost it your whole set would be corrupted.

You would NOT want to have different size devices/partitions in your RAID set because it stripes across all the drives equally. This means even if it were possible then you would never get all the space of larger drives/partitions as it would be limited to the number of stripes that could be placed on the smallest drive/partition.

Thanks JLightner, that really clears it up well! I just went ahead and searched around for a spare scsi and found one that was as close to identical as it gets (for us I suppose).

Thanks,

Harlin

To follow up on jlightner’s comments, a linux software raid 5 set can use different sizes of drives, but you are limited to only using the smallest drive’s capacity (or the smallest partition available for the Raid) across the set.

But you will get the best performance if the drives’ performance and capacities are matched. Even having one lower performing drive in the array can cause a marked drop in overall performance.

I think in my case I was going to need to go with exact capacities because (as I failed to mention earlier), this was using a hardware based array as opposed to using the OS to create them.