Storage Administration - Oracle Exadata Recipes: A Problem-Solution Approach - page 246

How It Works
Mirroring protects data by storing copies of data across different disks, providing access to user data in the event of a
disk failure. When you create an ASM disk group, you specify a level of redundancy—this redundancy level dictates
how many copies of data are maintained:

• Normal redundancy = two-way mirror

• High redundancy = three-way mirror

• External redundancy = no ASM mirroring; ASM uses mirroring functionality in the storage

array/subsystem, if available, to provide protection

With Exadata, Oracle protects and mirrors storage exclusively with ASM normal or high redundancy; there is no
external redundancy alternative on Exadata.

Not ■ You can elect to configure aSm disk groups with no mirroring (that is, no redundancy), even on exadata, but this
practice is not recommended. disk drives are one of the hardware components that are typically most prone to failure.

The redundancy level controls how many disk failures are tolerated without ASM un-mounting the disk group or
losing data.
When ASM allocates an extent for a mirrored file, it allocates a primary copy and one or two mirror copies. Oracle
places the mirror copy on a disk that's part of a different failure group . Failure groups are where mirror copies are
stored. If you lost disks in either the primary location or failure group(s) (but not both or all), Oracle would continue
to operate normally on the surviving copy.
ASM doesn't mirror physical disks or LUNs like traditional RAID—it mirrors database extents . This is a very
important design aspect of ASM—with normal or high redundancy, extents are mirrored on sectors from disks on one
or more failure groups. These failure groups consist of disks different from the primary extents.
On Oracle Exadata, Oracle always places mirrored extents on grid disks located in a different storage server. This
provides the flexibility that if you lost an entire storage server, you would still have access to either the primary or
mirrored extents.

Not ■ this also means that you can reboot a storage cell, patch it, or otherwise take disks offline in a storage cell
without impacting storage availability.

When ASM disk group redundancy is normal or high, database extents in a file for the disk group are mirrored to
a failure group. As such, failure groups are defined at the ASM disk group level. Again, normal redundancy means two
copies of an extent are created, one for the primary extent and one in a failure group. Loss of either disk, but not both,
will provide uninterrupted availability. For high redundancy, three copies (one primary and two failure group copies)
are created. In this configuration, Oracle can tolerate two failed disks across multiple failure groups.
Each disk in a disk group can belong to one and only one failure group. Failure groups should all be the same size,
or availability is impacted. With normal redundancy, at least two failure groups are required; for high redundancy,
three failure groups are required. On Exadata, Oracle ASM creates a set of failure groups for each cell.
One thing that's interesting about Oracle ASM is that if insufficient disks are available to satisfy mirroring based
on a redundancy level, Oracle will do as much as it can to mirror and then subsequently allocate remaining mirrors
when sufficient disks are available.