With the introduction of the new balancing algorithm in 6.0.3 balance is determined by data size per (sharded) collection per shard. This had the knock-on effect of changing the steps to execute the commonly used tactic of splitting an empty collection to maximize throughput when inserting data into a cluster.

Customers should be provided a way to provide their desired split points for their empty collection and then have the appropriate ranges automatically migrated to the other shards of the cluster.