Hmmm. Thanks for that. Makes for interesting reading. So it seems as though today's internal RAID cards perform better or the same in RAID 5 as in RAID 10 for both read and write. That's quite useful considering the extra capacity of RAID 5.
Yes.
Recent PCIe based cards have invalidated the old adage that 10 is always better than 10. First with sequential throughputs (reads are more similar than the difference between writes), and more recently with relational databases.
RAID 5 has become far more attractive than when BAARF got started (so long as it's done in a hardware solution), due to lower cost alone (fewer disks to meet capacity requirements at a bare minimum). But the faster write performance on such cards makes it even more attractive than 10 for sequential usage, and the costs are the proverbial icing on the cake.
There are still instances with relational databases that 10 can still be a better performer, but it depends on the specific usage and requirements, and if the budget will allow for it if it 10 is the faster way to go (i.e. additional disks to meet the capacity requirement vs. what 5 can do on fewer disks).
But from a point of view of external NAS, benchmarks show that RAID 10 still outperforms RAID 5. Would you agree?
I'm not a fan of inexpensive ready-made NAS units, as some are software based (5 may be listed in the specifications, but is a mistake due to the lack of a solution to the write hole issue associated with parity based arrays). 10 would be the only way to go on such units. They're also slow. For similar money, you can build one (even a SAN) on a Linux based system that's faster and sufficiently reliable (i.e. software based and use the ZFS file system to build Z-RAID/Z-RAID2). The "cost" of course, is self-support when something goes wrong. But support from Data Robotics isn't that great either, so it could be considered a wash (warranty support may still be an advantage to a cheap ready-made NAS).
There are also inexpensive hardware controller chips (i.e. Oxford/PLX, JMicron,...), but they're not built for speed (found in some NAS units, and some multi-disk enclosures such as the
Qx2 from OWC <uses an Oxford 936>). For these, they'd need to be tested (hopefully, there's a review that tested the performance on all supported levels) in order to make a decision that's not blind (based on a few specs and cost alone, and an unknown otherwise).
In such units though, the likelihood of 10 being faster than 5 (ignoring capacity) is much higher than 5. RAID 5 still has the advantage of capacity though (i.e. using 4x 3TB disks, 10 = 6TB usable, 5 = 9TB usable). So it still comes down to trying to balance performance, capacity, and cost. Always a compromise...
