HP StorageWorks SAN Virtualization Services Platform Best Practices Guide (5697-0935, May 2011)

11 Asynchronous mirroring
Asynchronous mirroring is accomplished by sending the changes at regular intervals over iSCSI
(WAN) to another SVSP domain. The number of PiTs kept at the source and destination is
configurable, allowing a flexible framework to meet your needs. If the link is not sized appropriately,
it is possible the PiTs will queue, compromising the RPO or RTO. It is also possible that changes
in a given period will exceed the link bandwidth and those changes will continue to be pushed
across the link. If this occurs, the time to complete the current PiT will be extended until the previous
copy is complete. Therefore, matching the rate of change to the bandwidth is important to the
success of asynchronous mirrors. To determine the rate of change, HP recommends that you set
up alerts to monitor it at frequent intervals. Once the rate of change is established, you can size
the link according to the RPO or RTO.
For many applications that run large I/O, or many I/Os to a MB segment, there is a minor
performance impact, but PiT usage may produce a bottleneck when the PiT is first created and has
a workload of high random writes with small request sizes. Although SVSP uses redirect-on-write
snapshots (PITs), it still must perform a segment read followed by a segment write to composite the
new data into a segment on first write after the snapshot (PITs) is created. A small write becomes
a 1 MB read, a move of the new data into the buffer, and a write of the buffer. Combined with a
desire for low RPOs that require high PIT rates and large segment sizes of SVSP of 1 MB per
segment, and it is likely that asynchronous mirroring will not perform as desired.
TIP: RPO consists of two measurements; the interval between consecutive PiTs and the transfer
time. The bandwidth affects how much data can be transferred in a given time, and since the next
transfer will not start until the previous transfer completes, it is important to size bandwidth correctly.
HP has an internal SVSP bandwidth tool that can determine the maximum copy rate for a specified
distance between two sites (the tool is often used to design SVSP Continuous Access solutions).
The maximum copy rate is used to determine the minimum copy time for a specified amount of
data. The copy rate can also be adjusted lower to help determine the most cost effective option.
See the HP StorageWorks SAN Virtualization Services Platform Administrator Guide for details
about site recovery failover with asynchronous mirrors.
Deploying asynchronous mirroring
A useful method for deploying asynchronous mirroring over a WAN is to bring the new equipment
for the disaster recovery site to the local site, create the new SVSP domain there, and connect the
SVSP domains. You can then create the asynchronous mirror tasks and allow them to run the initial
build over the fast and low latency local links. When the initial build is complete, you can suspend
the tasks, disconnect the SVSP domains, move the disaster recovery SVSP domain to its new physical
destination, reconnect the SVSP domains over iSCSI network, and resume the mirror tasks.
Using virtual disk groups
To minimize the impact of a virtual disk failure from affecting more than one virtual disk, use virtual
disk groups only when there is a requirement for data to be consistent on multiple virtual disks.
Examples of this requirement include:
• Ensuring data consistency between a database and transaction log virtual disks
• Using HP-UX Logical Volume Manager or similar applications
• Having a VMware ESX server that spreads a VMFS data store across multiple virtual disks
Deploying asynchronous mirroring 49