Avoiding Parity Limitations of RAID with RAID 10

RAID 10, also called RAID 0+1, is not one of the original defined RAID levels, but it is extremely effective and in many ways is superior to RAID 5, except for the capacity required. RAID 10 combines the mirroring of RAID 1 with the striping of RAID 0.

In essence, RAID 10 mirrors every member in a RAID 0 array. Thus, RAID 10 removes the RAID 5 write penalty from the equation and also provides much more redundancy depth than RAID 5. A RAID 10 array can lose more than two members and still continue to operate, as long as it does not lose both pairs of a mirrored member. Figure 9-6 shows a RAID 10 array made up of four mirrored pairs.

Figure 9-6. A RAID 10 Array with Four Mirrored Pairs

There is also no loss of performance while operating in degraded mode, because there is no parity to calculate on the fly. Instead, the data is simply read from the remaining member in the pair. More importantly, parity rebuilds are avoided when repopulating a replacement member in an arraya relatively simple disk remirror operation is performed instead. The lack of parity also allows RAID 10 arrays to scale to include many more members than RAID 5. In fact, there are no immediate scalability limits with RAID 10. Reliability does not decrease as members are added. Parity data reconstruction processes that become more complicated as the number of members increases are avoided altogether.

The problem with RAID 10 is that it has a 50% redundancy overhead compared to RAID 5, which can be less than 10% in arrays with more than ten members.

NOTE

With disk drives being some of the least expensive elements in the data center, I'm not sure that the cost difference of capacity between RAID 10 and RAID 5 can seriously be justified as a reason for using RAID 5 instead of RAID 10. Some people have told me they still prefer RAID 5 because it can have a smaller footprint than RAID 10. Understanding that footprint actually is important to data center management, the question is one of priority: do you need to optimize the physical aspects of your data center at the expense of performance, flexibility, and redundancy?