Tuning scheduled VM backups for Windows Server 2012 and 2012 R2 clusters

Beginning with Data Protection for Microsoft Hyper-V Version 8.1.2, you can back up more virtual machines (VMs) in parallel and across nodes in a cluster. A cluster node backup operation always retries the snapshot on volumes with snapshots that failed with a recoverable condition. You can also tune the number of VMs in a snapshot to reduce the workload of a snapshot for the Hyper-V host.

You can use the following options to tune how snapshots are taken during the backup:
  • Use the vmmaxparallel option to control how many VMs are sent in parallel to the IBM Spectrum Protect server. The setting for this option has the most notable impact on performance.
  • Use the vmmaxpersnapshot option to control how many VMs can be included in each snapshot that is created during the backup operation.

Before you back up a cluster, review and tune the values for these two options for the environment.

Use the following general approach to tune your cluster backup operations:
  1. Plan to use an appropriately sized and configured IBM Spectrum Protect server that uses container pools. For information about how to size the server, see IBM Spectrum Protect Blueprints.
  2. As a starting point, use the default values for the vmmaxpersnapshot and vmmaxparallel options.
  3. Run the backup schedule and note the results, such as whether backups completed within the schedule window or whether too many snapshot retries occurred.
  4. Adjust the value for the vmmaxparallel option to work in your environment. For example, set the value to 10.
  5. Adjust the value of vmmaxpersnapshot to a value that minimizes the number of retries that occur. The retries are reported in the backup statistics.

    When you choose a smaller number of VMs per snapshot, you increase the number of snapshots that are needed to complete a backup operation. This increase in snapshots can lead to delays during cluster backup operations of VMs on CSVs. The delay occurs because only one snapshot can be created at a time, and backup operations of other nodes in the schedule are delayed during snapshot creation. By increasing the number of VMs in a snapshot, you can reduce the number of snapshots that are taken for a backup operation.

    To determine the number of VMs to include in a snapshot, consider the following factors:
    • A snapshot with more VMs takes longer to complete and increases the load on the system. A larger number of VMs means that the snapshot persists longer, which can affect system performance.
    • The vmmaxpersnapshot and vmmaxparallel options work together to determine how many snapshots are taken in a backup operation. The vmmaxparallel option specifies how many VMs can be backed up simultaneously. Data Protection for Microsoft Hyper-V takes as many snapshots as needed to meet the vmmaxparallel setting.

      VMs are sorted and selected based on the volumes that are needed to create the snapshot for the VMs. A snapshot is created for a set of VMs that share a set of volumes. Thus, the number of snapshots varies depending upon the volumes that are used by the VMs. The number of VMs per snapshot never exceeds the value for the vmmaxpershapshot option.

      The following table shows examples of how many VMs can be processed per snapshot with various vmmaxpersnapshot and vmmaxparallel settings. In these examples, assume that all the VMs are on the same volume.
      Table 1. Number of snapshots and VMs (on the same volume) processed with the vmmaxpersnapshot and vmmaxparallel settings
      vmmaxpersnapshot setting vmmaxparallel setting Number of snapshots created
      10 20 Two snapshots are created with 10 VMs each. When the number of VMs being processed is less than the vmmaxparallel setting, another snapshot is taken.
      20 20 One snapshot is created containing 20 VMs.
      20 10 One snapshot is created containing 20 VMs, and 10 VMs are backed up due to the vmmaxparallel setting during the first run. The remaining 10 VMs are backed up during the second run (a second snapshot is not needed).

You can also use the vmmaxsnapshotretry option to specify the maximum number of times to retry a snapshot operation of a VM if the initial snapshot fails with a recoverable condition.