Loading...

Best Practices for the Virtual Server Agent with VMware

Table of Contents

Releasing Virtual Server Agent License

Using the Default Subclient

Include Multiple Virtual Machines for a Subclient

SAN Deployments

HotAdd Deployments

Activating Changed Block Tracking for Backups

Configuration Considerations

Location of Job Results Directory

Name of NFS Datastore

Issues with Application-Consistent Quiescing When a Virtual Machine Has More Than 7 Disks Attached to a Single SCSI Controller

Defining Rules Based on Datastore and Cluster Affinity to Discover Virtual Machines

Configure Separate Virtual Clients for vCenter and vCloud

Disable DRS for VSA Proxies on VMs

Releasing Virtual Server Agent License

If the Virtual Server Agent is no longer required to run data protection operations, you can release the Virtual Server Agent license instead of uninstalling the agent. If the agent is uninstalled, you will not be able to:

  • Perform Live Browse.
  • Unmount snapshots mounted by any Virtual Server Agent.

Using the Default Subclient

Use and run the default subclient regularly because it performs VM discovery operations.

The default subclient is used regularly in most deployments, but the following exceptions apply:

  • Multiple Virtual Server Agents are deployed in the environment and the 'Master' model is in use.
  • Multiple storage policies are required for different virtual machines.
  • Rule-based discovery is in use and different subclients are used to target a particular affinity.

Include Multiple Virtual Machines for a Subclient

The virtual machines in a subclient are all backed up in a single backup job. Generally the best practice is to include multiple virtual machines in each subclient. You can create multiple subclients to meet the following requirements:

  • Manage different classes of virtual machines.
  • Group virtual machines that reside on the same datastore or datastore cluster.
  • Group virtual machines that have the same storage policy and use the same backup method.
  • Stagger backup schedules across the enterprise. For more details, see the information about Jobs Stagger Time in Job Priorities and Priority Precedence.

Having a large number of subclients is not required. Creating one subclient for each virtual machine can cause scheduling problems, and can become unmanageable in larger environments.

A more useful strategy is to have different subclients that address different datastores. This approach enables backup jobs to run in parallel for better performance when backing up many virtual machines.

SAN Deployments

  • On the server hosting the Virtual Server Agent, ensure that automount has been disabled to disable automatic drive letter assignment:

    diskpart> automount disable

  • Scrub automount settings to ensure that previous drive letter assignments and other volume mount settings are cleared:

    diskpart> automount scrub

  • Zone the LUNs containing VMFS datastores so that the physical server hosting the Virtual Server Agent and MediaAgent can see and access the LUNs.
  • LUNs can be presented with read only access, but this setting will prevent SAN mode restores.
  • Do not initialize, write a signature to, or assign a drive letter to the VMFS volumes presented to the physical Windows host.
  • Do not attempt to mount, partition, or format the VMFS volumes presented to the physical Windows host.

HotAdd Deployments

  • When deploying the Virtual Server Agent for HotAdd Mode backups, choose the datastore with the largest VMFS block size to ensure backups can mount and back up virtual machines residing on all datastores.
  • On the server hosting the Virtual Server Agent, ensure that automount has been disabled to disable automatic drive letter assignment:

    diskpart> automount disable

  • Scrub automount settings to ensure that previous drive letter assignments and other volume mount settings are cleared:

    diskpart> automount scrub

Activating Changed Block Tracking for Backups

When Changed Block Tracking (CBT) is unavailable, backups revert to Cyclic Redundancy Check (CRC) to determine changed blocks. Since the Virtual Server Agent needs to read the entire virtual machine disk, CRC incremental backups may take almost as long as full backups even though the amount of data transferred and stored by an incremental backup is limited to the changed blocks within the virtual disk. For that reason, you should correct CBT on the problematic virtual machine as quickly as possible to take full advantage of VADP-based incremental backups.

Tools are available to check the correct installation of the VDDK and to verify and correct CBT issues. Contact your software provider for information on obtaining these tools.

Configuration Considerations

  • Regular backups should be performed on the default subclient. This will ensure new virtual machines are discovered on a regular basis along with the backup operation.
  • Adding one virtual machine per subclient is not recommended. This creates issues around scheduling and becomes unmanageable in larger environments.
  • Typically only the default Subclient will be used in most deployments except in the following cases:
    • Multiple Storage Policies are required for different virtual machines.
    • Rule-based discovery is in use and different subclients will be used to target a particular affinity.

Location of Job Results Directory

The Job Results directory stores the job results files from backup and restore operations of a client.

The user impersonation account specified for the job results directory will take precedence and will be used to backup and restore data from a virtual machine. This may result into file access related issues during the backup. Therefore, it is recommended to use a local folder on the client computer as the job results directory.

Name of NFS Datastore

It is recommended to use a short name for an NFS datastore. When you perform the IntelliSnap backup of an NFS datastore, Commvault can append up to 20 characters with the name of NFS datastore to create the volume label. The ESX server supports a volume label of 42 characters.

Issues with Application-Consistent Quiescing When a Virtual Machine Has More Than 7 Disks Attached to a Single SCSI Controller

There are known issues when attempting application-consistent snapshots against virtual machines that have more than seven disks and are connected to a single SCSI controller. This can be resolved by adding additional controllers or disabling application-consistent quiescing.

For more information, see the following VMware knowledge base article:

Snapshot of a Windows server fails with the error: Protocol error from VMX (1037071).

Defining Rules Based on Datastore and Cluster Affinity to Discover Virtual Machines

When you have a large number of datastores, you should define rules based on cluster affinity to discover virtual machines. This approach avoids the need for a large number of subclients and enables virtual machine protection operations to be evenly distributed across available proxies.

Configure Separate Virtual Clients for vCenter and vCloud

If a VMware instance is configured to access both vCenter and a vCloud Director, backup jobs fail to collect vCloud information for virtual machines in the vCenter. As a result, the backup job status shows "Completed with Errors" even though virtual machines in vCenter and vCloud are actually backed up.

To prevent this message from being generated, use the CommCell Console to configure separate virtual clients for vCloud and vSphere.

Disable DRS for VSA Proxies on VMs

If the Virtual Server Agent is installed on a VM, disable Distributed Resource Scheduler (DRS) for the VM in the VM overrides page in vSphere.

If DRS is enabled for a VSA proxy, VMware can move (vMotion) the VM to a different host during a backup operation, disrupting the operation. A vMotion of a VSA proxy during a backup operation can also cause other issues, such as causing subsequent backup operations to fail, causing the VSA proxy to become unresponsive, leaving HotAdd disks mounted, and requiring manual cleanup.

Last modified: 2/23/2018 8:41:52 PM