We have a three node Hyper-V cluster with shared iscsi storage. Each node has four connections to the storage and other different connections for clustering, management access and networking.
While running our Veeam jobs, I've noticed that all I/O with the storage only happens on the node that owns the disks. So when the backup is running, the data stream is read from the storage by the node that owns the CSV disk where the VM files are stored, and transferred to the node that owns the VM through the clustering network.
This is obviously a bottleneck and while it was easy to notice during backup because of the continuous data stream, the same happens for all application workloads; we have about 40 VMs on the cluster.
What is your opinion on this? Is that normal behavior or may it be caused by some wrong configuration?
Any suggestions?
Thanks in advance, Alessandro
Alessandro