WAN Accelerator Design Questions

[tbrownfield]

We have multiple datacenters around the country and are moving to a mesh design for our backups. The goal is to store 7 days of backups local to each site and then have backup copy jobs store 45 days at a different datacenter. Questions:

1. I see that WAN acceleration is recommended for links less than 100Mbs. Would it still be recommended in order to reduce WAN saturation between sites so that production traffic isn't impacted? Several of our sites have large pipes but we will be backing up hundreds of VMs from one site to another across the same link. (I don't want to have to confess to my boss that I'm the one that took site B down in the middle of the night!)
2. In our design, site A will send backup copies of local VMs to site B and site B will also send backup copies of its local VMs to site A. Can the WAN accelerator in each site act as both a target and source or should we split these roles out for performance so that the target global cache and source cache usage don't compete with each other? Some sites we have SSD storage we can use but not all. Again, we're talking hundreds of VMs and multiple TB of data.
3. Can someone remind me what the optimal number of VMs per backup job is and is there a preferred way to divide them up? For example, will it be better to place all 2012 R2 VMs in one job, 2016 in another, etc. or does it matter?

Thanks!

[Shestakov]

1. Yes, WAN accelerator will still be useful.
2. You can do that. Do you consider sending traffic from A to B and in the opposite direction at the same time?
3. There is no recommended number. VBR works fine with big number(100+) of VMs in one job so big customers go for it for reasons of convenience. You will get better deduplication results if put VMs with same OS/applications under the job.
How many VMs do you have? Do you use per-VM backup option on repository?
Thanks!