Databases Reference
In-Depth Information
RAID level 0
This is the simplest RAID level, and it's not recommended to use it in any production
environment by Oracle itself.
In this RAID level only non-redundant striping is implemented, that is, the data is striped across
multiple disks to obtain better read/write performance. However, this is a non-redundant
solution, so any disk failure causes the outage of the entire array and hence data loss. It's very
cheap because it doesn't need more storage capacity than the actual space required.
RAID level 1
RAID level 1 implements a disk mirroring strategy. For each disk drive there is at least one
identical disk drive on which an exact copy of the data is maintained. There can be
n-1
outages, where
n
is the number of drives on which the data is simultaneously stored, without
any data loss. If hot-swappable drives are used there could be no application outage even
after
n-1
outages in different disks.
The main defect of this implementation is the cost per megabyte, because for each megabyte
of data stored we need
n
megabytes of storage to ensure the design redundancy.
Performance, as compared to the performance of a single drive, is slightly better in read times if
the controller can choose the disk from which to read the data considering the least I/O cost.
We can use RAID level 1 to store control files, system tablespace, flashback recovery area,
undo segments, and data files. It's not a good idea to store redo logs and temporary segments
on RAID level 1 configurations, because redo logs should be multiplexed on different disks,
and there is no need to protect temporary segments from failures.
RAID level 5
RAID level 5 introduces the concepts of Block-interleave and distributed parity.
In a RAID 3 array, there is one disk dedicated to the storage of parity data for the other disks
in the array, and data is striped using a strip of 1 bit. RAID 5 introduces two variations of
the schema—stripes size is configurable and parity data is not stored on a single drive but is
distributed among all disks of the array.
The reason to choose RAID level 5 is to obtain a redundant storage solution, cheaper than
RAID level 1. Due to storage of parity data, however, write operation performance suffers. This
is because, when writing some data, we need to read the old data and parity value, and write
the new data and parity value, resulting in four I/O operations.
Read performance is excellent when the data fits in a single striping segment, allowing a
heavy concurrency on data. Except undo segments and intensive-write data files, all other
Oracle files can be stored in a RAID level 5 array—when you cannot use RAID level 0+1,
especially with read-only data files.
