Pretty high activity on node after migration

So, SCSI (scsi2) with the default controller (LSI 53C895A). I guess this should be good enough (though changing to VirtIO SCSI would reduce CPU load). No other ideas, sorry. You can surely play with VM settings, maybe you’ll find something interesting.

Regarding RAM, I was running 12 TB worth of nodes on a 2GB RAM machine and it was good enough. Slowish, but never had a problem like yours.

There are circumstances where two file walker processes (one doing GC, one doing space calculation) compete for resources. You may disable the latter and check whether it helps.