So if I understand this correctly, if I have a VM that is 200GB in size, with off-host backup, the 200GB will be moved to the proxy machine. From there the proxy will deduplicate and compress the bits. Now probably at much smaller size (let's say 50GB) the data is moved from the proxy to my NAS. That is 250GB of network traffic.
With on-host backup, all the processing occurs at the host, so only 50GB of total data will be moved on the network from the host to the NAS. Now this seems like a good benefit if the processing on the host happens in the host itself not the VM guest.
With the on-host approach where does the processing (deduplicate, compress, etc.) occur? On the hyper-V host itself or on the guest OS? In other words will the logical CPUs and memory assigned to the VM, being backed up, be used or the CPU and memory left over to the host?
It seems to me that if the host resources and not the VM resources are used for the on-host backup approach then the users of the VM will see no performance degradation while the backup is occurring and will also benefit from less network traffic.
Unless I am missing something, my conclusion is that in my scenario, off-host backup offers less benefit than on-host as long as enough CPU and memory is left to the Hyper-V host to do the processing since those resources are segregated from the actual VMs. The only resource shared between the host and the VMs is local disk access. I don't know how much of a problem that is though. Is moving the data from the host to the proxy more or less taxing on disk access than doing the deduplication and compression?