Database Reference
In-Depth Information
protection mechanism, it's important that you understand how it works. The
advantages can be significantly higher performance during failure,
significantly faster recovery from failure, and greater predictability.
RAID Penalties
Random IO patterns, read/write bias, and failure events have a big impact on
performance due to the overheads and penalties for read and write operations
associated with using RAID. This is especially so with spinning disks. Storage array
vendors have come up with many ways to try and work around some of the limitations
with RAID, including the smart use of read and write caches. In your storage design,
though, we recommend you plan your performance based on the physical characteristics
of the underlying disks and plan for the rest to be a bonus.
Table 6.10
displays the IO
penalties during normal operations for each of the common RAID schemes.
Table 6.10
RAID IO Penalties During Normal Operations
IOPS listed in
Table 6.10
are per disk. RAID 0 is included for illustrative purposes
only and is not recommended, as it is simple disk striping with no data protection.
Note
The basis for the IOPS calculations in
Table 6.10
is the rotational latency
and average seek time of each disk. These will be different depending on
the latency characteristics of different manufacturers' disks. This would
also not apply for solid state disks and PCIe NAND flash devices. For
further information about IOPS, see
http://en.wikipedia.org/wiki/IOPS
and
As you can see from
Table 6.10
, if you have a very write-biased workload, you could
get very low effective IOPS from your RAID disks. This is the primary reason why
arrays have write cache—and in some cases, lots of it. This allows the array to offset
