XDCR resource consumption

We are setting up XDCR from CB3 cluster to CB4 cluster in order to upgrade. We have one bucket to replicate.
We are considering using the default configuration of XDCR because we do not want to over utilize our source machines.
While using the default configuration, how much RAM, CPU and I/O should be “reserved” for the XDCR (in both initial replication and ongoing replication)? I saw in one blog that they recommend having extra 2GB of RAM for the XDCR, but what about CPU and I/O? we wouldn’t want our production DBs to be influenced too much from this operation.

Another question - will the initial replication (from zero to full) will consume more resources than the ongoing (delta catch ups)