Only other comment/suggestion is to experiment with different LACP LAG's between the stack and the MLAG core. We've found that L3_L4 seems to provide a bit better distribution and balances the links better in most of our cases (typical traffic, mix of voice/data).