Symmetric IRB and VLAN everywhere
In my reading, one of the biggest points against asymmetric IRB is that all SVIs need to be defined on all VTEPs. The books go on to say that symmetric IRB is better because you have a L3 VNI (for each VRF) that is shared across all VTEPs allowing them to not need to have a destination SVI.
However, if you are a large ESX shop (for example) you would need to have most VLANs in all VTEPs to allow for VM migrations. In this case, isn't symmetric IRB actually worse because in addition to all the VLANs (and all the arp and mac address tables) you also have to maintain the L3 VNI information as well? How could any single switch keep up with an entire data centers arp and mac address tables? Is there some kind of optimization that I am missing here, or does it really come down to managing which VLANs go where in your data center?