HP P6000 Continuous Access Implementation Guide (T3680-96431, August 2012)

snapshot, mirrorclone, or snapclone feature to create a copy of the destination virtual disk before
a normalization starts. If a major failure occurs at the local site during a normalization, HP P6000
Business Copy can provide a clean copy of data as it existed before normalization writes started
on the remote array. (Note that any new writes that occurred on the source between the time the
copy was created and the major failure occurred are lost.)
It is recommended that you use the HP P6000 Business Copy mirrorclone feature if supported on
your XCS version, or a fully allocated snapshot if your XCS version supports instant restore from
snapshot. If neither mirrorclone nor snapshot is available, use the HP P6000 Business Copy
snapclone feature. As a best practice, whenever a link is expected to be down more than several
minutes, create a copy of the destination virtual disk.
Some versions of controller software support instant restore of a snapshot of a mirrorclone to the
mirrorclone parent. When a DR group source member is instantly restored from a snapshot (either
its own snapshot or the snapshot of an attached mirrorclone), the group will use the available bit
map to copy only blocks that have changed, rather than the entire content of the DR group. This
is more efficient than performing a full copy of the data and is referred to as a “fast copy”.
The fast copy functionality is only associated with the source DR group member. The bit map used
for the fast copy is only stored on the source DR group member's array. The bit map is not available
on the destination DR group member’s array, so an asynchronous DR group cannot initiate a fast
copy after a failover. A full copy is required to ensure the data on the new source is copied to the
new destination. The new source will create and maintain a bit map for fast copy should a
communication failure or group suspend occur. See the HP P6000 Enterprise Virtual Array
Compatibility Reference to determine if your array controller software supports this feature.
NOTE: You cannot use these procedures if a normalization is already in process.
1. Navigate to each affected DR group and suspend replication.
2. Make a snapclone of the destination virtual disk using the procedures described in the
application's online help. On versions of XCS controller software that support
reverse-resynchronization, HP recommends using mirrorclones or fully allocated snapshots.
3. Use HP P6000 Replication Solutions Manager to navigate to each affected DR group and
resume replication. Replication will resume when the links are restored.
NOTE: A DR group must be suspended to perform an instant restore from a mirrorclone, a
snapshot of a mirrorclone, or a snapshot of a virtual disk.
Optimizing performance
The following information describes how you can improve performance when using HP P6000
Continuous Access.
Load balancing
NOTE: The following recommendations provide a fully redundant solution that will survive an ISL
failure and continue to provide full bandwidth performance.
For best performance of remote replication, the average workload (reads and writes) should be
equal on both controllers and on both fabrics and ISLs. To obtain this balance, take measurements
and keep the utilization rate of each ISL below 40 percent. If one link fails, the average utilization
of the operating link should not exceed 80 percent. Similarly, the utilization rate of a single controller
as measured on all host ports should not exceed 45 percent on average, or peak above 50 percent,
to prevent overloading the surviving controller should one controller fail.
In general, let the hosts manage I/O load balancing. Remember, all members of a DR group must
share a preferred controller, therefore load balancing a single application with all of its virtual
disks in a single DR group (as required) across multiple controllers is not possible.
Optimizing performance 125