Brocade Communications Systems StorageWorks 4400 - Enterprise Virtual Array Instrukcja Użytkownika Pobierz pdf (Strona 32)

The impact of test failovers that include application recovery can be mitigated with adequate disk array

sizing. These additional IOPS must be factored in when sizing the EVA disk group to allow transparent

failover testing during production hours. When deploying Site Recovery Manager in an active/passive

datacenter scenario, the impact of test failover on I/O performance is negligible. Figure 20 shows a

slowdown of an Exchange 2007 mailbox server running at site A, caused by the test of a recovery plan for

a SQL server application located at Site B.

Figure 20. Limited performance impact caused by a test failover

In our test environment, the performance impact is limited in both intensity and duration. This is because

there is sufficient headroom on the array as we include some margin in the sizing exercise to support

application recovery testing and accommodate for future growth. However, smaller configurations with less

physical drives can exhibit more important disruption.

Networking consideration

Some applications have a strong dependency on infrastructure components, such as DHCP, DNS, or Active

Directory domain controllers (DC). These components often use different protection techniques to provide

resiliency and may not be part of a Site Recovery Manager recovery plan. This can preclude a successful

execution of the plan from an application perspective, especially when using the bubble virtual network. For

example, failing over protected Exchange 2007 mailbox servers requires domain controller availability. If

DC is not available in the isolated network environment, the mailbox servers will not start successfully and

cannot mount the mailbox databases.

Recovery plan and failover considerations

Once started, recovery plans have minimal interaction capabilities. Therefore, it is important to use Site

Recovery Manager test functionality to practice and validate the plan prior to execution. Doing so ensures

proper startup order of VMs and application dependencies. You can (and should) create multiple recovery

plans that are based on application tiering and dependency. Multiple recovery plans can be started

simultaneously to bring up the environment faster after a disaster occurs.

1 2 ... 27 28 29 30 31 32 33 34 35 36 37 38

Brak uwag

Brocade Communications Systems StorageWorks 4400 - Enterprise Virtual Array Instrukcja Użytkownika Strona 32