RAID5 SATA rebuild times

Hi, does anyone have any practical experience with rebuild times of
SATA arrays? I've heard many recommendations suggesting that RAID5
SATA arrays with big (750GB, 1TB) disks should not have too many
spindles due to long rebuild times - but how many spindles is 'too
many'?

I currently work on the basis of anything between 4+1 to 7+1 and it
would be nice to have some evidence to support this. RAID6 would help
but that's not an option unfortunately.

Thanks
Jono

On Feb 7, 2:30 pm, Jono968 wrote:
Hi, does anyone have any practical experience with rebuild times of
SATA arrays? I've heard many recommendations suggesting that RAID5
SATA arrays with big (750GB, 1TB) disks should not have too many
spindles due to long rebuild times - but how many spindles is 'too
many'?

I just finished a test early this morning on a Nexsan SATABoy
containing a 13-member RAID-6 set consisting of 750GB drives (with 1
hot spare in the 14-drive enclosure). We yanked a drive out and
watched the rebuild - it took 18.5 hours. There wasn't much load on
the array during this time and the rebuild priority was set to
"medium" - 3 on a scale of 1 to 5.

How many spindles is "too many" depends on how you define "long
rebuild times" and what level of risk you want to have while the
raidset is rebuilding. Personally, I think that 18.5 hours is pretty
darn good for a 7.5TB raidset.

I did not do a test of a rebuild on a RAID5 set on this array.

.../Ed

Jono968 wrote:
Hi, does anyone have any practical experience with rebuild times of
SATA arrays? I've heard many recommendations suggesting that RAID5
SATA arrays with big (750GB, 1TB) disks should not have too many
spindles due to long rebuild times - but how many spindles is 'too
many'?

I currently work on the basis of anything between 4+1 to 7+1 and it
would be nice to have some evidence to support this. RAID6 would help
but that's not an option unfortunately.

The rebuild time of a SATA or SCSI or FC RAID5 array is as dependent on
the raid controller as it is on number of spindles in the array and the
trade-off between performance hit during the rebuild and the desire to
restore redundancy.