With 4+1 the Clariion can perform full stripe writes, while with 14+1 this is almost impossible. A 4+1 Raid Group has the capacity of 4 data disks, but performs like 5, while a 14+1 has 14 data disks and performs like 15. Needlessly to say that the 4+1 is relatively better.
Furthermore a rebuild to a hot spare in a 4+1 only needs to read data from 4 disks to be able to rebuild the data where a rebuild on the 14+1 needs 14 drives to rebuild number 15. So here the 4+1 wins again.
Now you might say: 15 disks outperform a 4+1. This is true, but I would compare a 14+1 to three 4+1 RGs. If you then create LUNs on the (4+1)s, use METALUNs with 3 components, so you'll have the performance of 15 drives on the LUN.
The only reason why one would use the 14+1 is perhaps cost. You only have 1 drive for parity, which is cheaper than 3 if you decided to create 3 RGs of (4+1) out of the 15 drives. Plus 14+1 gives you the capacity of 14 drives where three times a (4+1) only gives you 3x4=12 drives of capacity.
EMC calls it (4+1) RG the performance RG and a (8+1) the poor mans RG. (14+1) is even worse....
As RRR wrote your random writes are going to be adversely affected because of the very large stripe size and any rebuilds will be extended.
EMC Unified Storage System Fundamentals in its "RAID group' section has an explanation of how stripe size affects performance. In addition, its 'Rebuilds' section describes the affects of very large RAID groups on rebuild time, which also affects performance. Fundamentals is available on Powerlink.
RRR
6 Operator
•
5739 Posts
1052
0
Posted November 24th, 2011 03:00
With 4+1 the Clariion can perform full stripe writes, while with 14+1 this is almost impossible. A 4+1 Raid Group has the capacity of 4 data disks, but performs like 5, while a 14+1 has 14 data disks and performs like 15. Needlessly to say that the 4+1 is relatively better.
Furthermore a rebuild to a hot spare in a 4+1 only needs to read data from 4 disks to be able to rebuild the data where a rebuild on the 14+1 needs 14 drives to rebuild number 15. So here the 4+1 wins again.
Now you might say: 15 disks outperform a 4+1. This is true, but I would compare a 14+1 to three 4+1 RGs. If you then create LUNs on the (4+1)s, use METALUNs with 3 components, so you'll have the performance of 15 drives on the LUN.
The only reason why one would use the 14+1 is perhaps cost. You only have 1 drive for parity, which is cheaper than 3 if you decided to create 3 RGs of (4+1) out of the 15 drives. Plus 14+1 gives you the capacity of 14 drives where three times a (4+1) only gives you 3x4=12 drives of capacity.
EMC calls it (4+1) RG the performance RG and a (8+1) the poor mans RG. (14+1) is even worse....