Problem in some environments using Backup Exec to backup Hyper-V CSV (Cluster Shared Volumes) virtual machines (VM’s) the performance of the backup can be from none to a small degradation to a more significant difference in throughput when the VM are owned by one node and the CSV coordinator role is owned by the node the backup is done through Cause Depending on the Hyper-V CSV configuration this may have a negligible or a more noticeable difference. When the backup processes the CSV VM’s it will use a redirect mode for those VM’s that are not owned by the node that is the owner of the CSV. As an example; the volumes used in the CSV are SAN based but when the backups are running the the data transfer will be using redirect mode (for CSV v3 the state of the CSV will still state online during backups where CSV 2 during backups the state will have changed to redirected) . This will mean that it will use the cluster network with the lowest metric between the nodes for this backup data transfer using SMB and not directly from the SAN. The node that has the cluster coordinator role will be the only node that has direct access to the SAN storage and as such backups of VM’s that are controlled by this node will have the fastest throughput. For Windows 2008 R2 nodes the cluster redirect using the cluster network could only use one physical nic, but from Windows 2012 and later this was further improved in Windows 2012 R2 it is using SMB Multichannel (SMB2). To find out if if this smb redirect access is using the correct networks find the metric that is assigned to the networks used in the cluster A high performing network will show very little performance impact as opposed to a lesser performing network.
Solution Improve performance by upgrading the Hyper-V cluster network and/or make sure it is configured to use the correct cluster network(s)