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

20-millisecond average response time budget for latency sensitive applications such as Microsoft
Exchange.
Sizing bandwidth for asynchronous replication
HP P6000 Continuous Access includes an enhanced buffering technique for asynchronous replication
solutions that utilizes the DR group write history log. This disk-based journaling insulates application
users from latency delays caused by propagation delays and intermittent congestion on the link.
However, enhanced asynchronous replication can leave user data can be at risk from a site disaster.
The capacity of the link to move data from the log file determines the amount of exposure.
Establishing an optimal balance between the cost of bandwidth and the value of the data being
protected requires an accurate sizing.
It may appear that sizing link capacity equal to the average write byte rate is optimal. After all,
whatever data goes into the log must be replicated. However, there are several problems in using
averages. The primary issue is that averaging fails to take into account the tolerance to data loss
as specified in the RPO. The second is a practical matter when computing averages. What time
interval is meaningful for the averaging? While it may be convenient to look at the average change
rate because this information is often readily available, using averages will usually lead to a
sub-optimal or undersized bandwidth capacity.
Evaluating bandwidth capacity
Compare the average load and peak write rate of your applications with the capacity of intersite
link technologies and determine which technology is most effective. With XCS 6.000 and later,
the maximum capacity measure is averaged over the RPO interval. This limitation allows I/O from
a failed link or fabric to run on the active link or fabric without additional failures caused by
overloading the surviving fabric.
You can use the Data Replication Designer to analyze bandwidth data. The Data Replication
Designer is supported on Windows only. See “Tools for gathering SAN data” (page 14) for more
information on Data Replication Designer and other tools available for evaluating bandwidth.
Choosing a write mode
You specify the replication write mode when you create DR groups. The choice of write mode,
which is a business decision, has implications for bandwidth requirements and RPO. Synchronous
mode provides greater data currency because RPO will be zero. Asynchronous mode provides
faster response to server I/O, but at the risk of losing data queued at the source side if a site
disaster occurs. Asynchronous write mode can be basic or enhanced, depending on the software
version of the controller.
The write mode selection has implications for the bandwidth required for the intersite link. In general,
synchronous mode (and shorter RPOs) requires higher bandwidth and smaller network latencies.
For instance, synchronous mode can require twice the bandwidth during average workloads and
ten times the bandwidth during peak loads.
For complete information on which write modes are supported on each version of controller software,
see the HP P6000 Enterprise Virtual Array Compatibility Reference. For more information about
RPO, see “Recovery point objective” (page 18).
20 Designing a remote application solution