Interested in your choice of Raft. How well does that work when you have heterogeneous machines with varying degrees of latency to each other?